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 if (git_check_attr(&the_index, path, check)) 974 return 0; 975 if (ATTR_FALSE(check->items[0].value)) 976 return 1; 977 return 0; 978} 979 980/* 981 * When adding an object, check whether we have already added it 982 * to our packing list. If so, we can skip. However, if we are 983 * being asked to excludei t, but the previous mention was to include 984 * it, make sure to adjust its flags and tweak our numbers accordingly. 985 * 986 * As an optimization, we pass out the index position where we would have 987 * found the item, since that saves us from having to look it up again a 988 * few lines later when we want to add the new entry. 989 */ 990static int have_duplicate_entry(const struct object_id *oid, 991 int exclude, 992 uint32_t *index_pos) 993{ 994 struct object_entry *entry; 995 996 entry = packlist_find(&to_pack, oid->hash, index_pos); 997 if (!entry) 998 return 0; 9991000 if (exclude) {1001 if (!entry->preferred_base)1002 nr_result--;1003 entry->preferred_base = 1;1004 }10051006 return 1;1007}10081009static int want_found_object(int exclude, struct packed_git *p)1010{1011 if (exclude)1012 return 1;1013 if (incremental)1014 return 0;10151016 /*1017 * When asked to do --local (do not include an object that appears in a1018 * pack we borrow from elsewhere) or --honor-pack-keep (do not include1019 * an object that appears in a pack marked with .keep), finding a pack1020 * that matches the criteria is sufficient for us to decide to omit it.1021 * However, even if this pack does not satisfy the criteria, we need to1022 * make sure no copy of this object appears in _any_ pack that makes us1023 * to omit the object, so we need to check all the packs.1024 *1025 * We can however first check whether these options can possible matter;1026 * if they do not matter we know we want the object in generated pack.1027 * Otherwise, we signal "-1" at the end to tell the caller that we do1028 * not know either way, and it needs to check more packs.1029 */1030 if (!ignore_packed_keep_on_disk &&1031 !ignore_packed_keep_in_core &&1032 (!local || !have_non_local_packs))1033 return 1;10341035 if (local && !p->pack_local)1036 return 0;1037 if (p->pack_local &&1038 ((ignore_packed_keep_on_disk && p->pack_keep) ||1039 (ignore_packed_keep_in_core && p->pack_keep_in_core)))1040 return 0;10411042 /* we don't know yet; keep looking for more packs */1043 return -1;1044}10451046/*1047 * Check whether we want the object in the pack (e.g., we do not want1048 * objects found in non-local stores if the "--local" option was used).1049 *1050 * If the caller already knows an existing pack it wants to take the object1051 * from, that is passed in *found_pack and *found_offset; otherwise this1052 * function finds if there is any pack that has the object and returns the pack1053 * and its offset in these variables.1054 */1055static int want_object_in_pack(const struct object_id *oid,1056 int exclude,1057 struct packed_git **found_pack,1058 off_t *found_offset)1059{1060 int want;1061 struct list_head *pos;1062 struct multi_pack_index *m;10631064 if (!exclude && local && has_loose_object_nonlocal(oid))1065 return 0;10661067 /*1068 * If we already know the pack object lives in, start checks from that1069 * pack - in the usual case when neither --local was given nor .keep files1070 * are present we will determine the answer right now.1071 */1072 if (*found_pack) {1073 want = want_found_object(exclude, *found_pack);1074 if (want != -1)1075 return want;1076 }10771078 for (m = get_multi_pack_index(the_repository); m; m = m->next) {1079 struct pack_entry e;1080 if (fill_midx_entry(oid, &e, m)) {1081 struct packed_git *p = e.p;1082 off_t offset;10831084 if (p == *found_pack)1085 offset = *found_offset;1086 else1087 offset = find_pack_entry_one(oid->hash, p);10881089 if (offset) {1090 if (!*found_pack) {1091 if (!is_pack_valid(p))1092 continue;1093 *found_offset = offset;1094 *found_pack = p;1095 }1096 want = want_found_object(exclude, p);1097 if (want != -1)1098 return want;1099 }1100 }1101 }11021103 list_for_each(pos, get_packed_git_mru(the_repository)) {1104 struct packed_git *p = list_entry(pos, struct packed_git, mru);1105 off_t offset;11061107 if (p == *found_pack)1108 offset = *found_offset;1109 else1110 offset = find_pack_entry_one(oid->hash, p);11111112 if (offset) {1113 if (!*found_pack) {1114 if (!is_pack_valid(p))1115 continue;1116 *found_offset = offset;1117 *found_pack = p;1118 }1119 want = want_found_object(exclude, p);1120 if (!exclude && want > 0)1121 list_move(&p->mru,1122 get_packed_git_mru(the_repository));1123 if (want != -1)1124 return want;1125 }1126 }11271128 return 1;1129}11301131static void create_object_entry(const struct object_id *oid,1132 enum object_type type,1133 uint32_t hash,1134 int exclude,1135 int no_try_delta,1136 uint32_t index_pos,1137 struct packed_git *found_pack,1138 off_t found_offset)1139{1140 struct object_entry *entry;11411142 entry = packlist_alloc(&to_pack, oid->hash, index_pos);1143 entry->hash = hash;1144 oe_set_type(entry, type);1145 if (exclude)1146 entry->preferred_base = 1;1147 else1148 nr_result++;1149 if (found_pack) {1150 oe_set_in_pack(&to_pack, entry, found_pack);1151 entry->in_pack_offset = found_offset;1152 }11531154 entry->no_try_delta = no_try_delta;1155}11561157static const char no_closure_warning[] = N_(1158"disabling bitmap writing, as some objects are not being packed"1159);11601161static int add_object_entry(const struct object_id *oid, enum object_type type,1162 const char *name, int exclude)1163{1164 struct packed_git *found_pack = NULL;1165 off_t found_offset = 0;1166 uint32_t index_pos;11671168 display_progress(progress_state, ++nr_seen);11691170 if (have_duplicate_entry(oid, exclude, &index_pos))1171 return 0;11721173 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1174 /* The pack is missing an object, so it will not have closure */1175 if (write_bitmap_index) {1176 warning(_(no_closure_warning));1177 write_bitmap_index = 0;1178 }1179 return 0;1180 }11811182 create_object_entry(oid, type, pack_name_hash(name),1183 exclude, name && no_try_delta(name),1184 index_pos, found_pack, found_offset);1185 return 1;1186}11871188static int add_object_entry_from_bitmap(const struct object_id *oid,1189 enum object_type type,1190 int flags, uint32_t name_hash,1191 struct packed_git *pack, off_t offset)1192{1193 uint32_t index_pos;11941195 display_progress(progress_state, ++nr_seen);11961197 if (have_duplicate_entry(oid, 0, &index_pos))1198 return 0;11991200 if (!want_object_in_pack(oid, 0, &pack, &offset))1201 return 0;12021203 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);1204 return 1;1205}12061207struct pbase_tree_cache {1208 struct object_id oid;1209 int ref;1210 int temporary;1211 void *tree_data;1212 unsigned long tree_size;1213};12141215static struct pbase_tree_cache *(pbase_tree_cache[256]);1216static int pbase_tree_cache_ix(const struct object_id *oid)1217{1218 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);1219}1220static int pbase_tree_cache_ix_incr(int ix)1221{1222 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);1223}12241225static struct pbase_tree {1226 struct pbase_tree *next;1227 /* This is a phony "cache" entry; we are not1228 * going to evict it or find it through _get()1229 * mechanism -- this is for the toplevel node that1230 * would almost always change with any commit.1231 */1232 struct pbase_tree_cache pcache;1233} *pbase_tree;12341235static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1236{1237 struct pbase_tree_cache *ent, *nent;1238 void *data;1239 unsigned long size;1240 enum object_type type;1241 int neigh;1242 int my_ix = pbase_tree_cache_ix(oid);1243 int available_ix = -1;12441245 /* pbase-tree-cache acts as a limited hashtable.1246 * your object will be found at your index or within a few1247 * slots after that slot if it is cached.1248 */1249 for (neigh = 0; neigh < 8; neigh++) {1250 ent = pbase_tree_cache[my_ix];1251 if (ent && !oidcmp(&ent->oid, oid)) {1252 ent->ref++;1253 return ent;1254 }1255 else if (((available_ix < 0) && (!ent || !ent->ref)) ||1256 ((0 <= available_ix) &&1257 (!ent && pbase_tree_cache[available_ix])))1258 available_ix = my_ix;1259 if (!ent)1260 break;1261 my_ix = pbase_tree_cache_ix_incr(my_ix);1262 }12631264 /* Did not find one. Either we got a bogus request or1265 * we need to read and perhaps cache.1266 */1267 data = read_object_file(oid, &type, &size);1268 if (!data)1269 return NULL;1270 if (type != OBJ_TREE) {1271 free(data);1272 return NULL;1273 }12741275 /* We need to either cache or return a throwaway copy */12761277 if (available_ix < 0)1278 ent = NULL;1279 else {1280 ent = pbase_tree_cache[available_ix];1281 my_ix = available_ix;1282 }12831284 if (!ent) {1285 nent = xmalloc(sizeof(*nent));1286 nent->temporary = (available_ix < 0);1287 }1288 else {1289 /* evict and reuse */1290 free(ent->tree_data);1291 nent = ent;1292 }1293 oidcpy(&nent->oid, oid);1294 nent->tree_data = data;1295 nent->tree_size = size;1296 nent->ref = 1;1297 if (!nent->temporary)1298 pbase_tree_cache[my_ix] = nent;1299 return nent;1300}13011302static void pbase_tree_put(struct pbase_tree_cache *cache)1303{1304 if (!cache->temporary) {1305 cache->ref--;1306 return;1307 }1308 free(cache->tree_data);1309 free(cache);1310}13111312static int name_cmp_len(const char *name)1313{1314 int i;1315 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)1316 ;1317 return i;1318}13191320static void add_pbase_object(struct tree_desc *tree,1321 const char *name,1322 int cmplen,1323 const char *fullname)1324{1325 struct name_entry entry;1326 int cmp;13271328 while (tree_entry(tree,&entry)) {1329 if (S_ISGITLINK(entry.mode))1330 continue;1331 cmp = tree_entry_len(&entry) != cmplen ? 1 :1332 memcmp(name, entry.path, cmplen);1333 if (cmp > 0)1334 continue;1335 if (cmp < 0)1336 return;1337 if (name[cmplen] != '/') {1338 add_object_entry(entry.oid,1339 object_type(entry.mode),1340 fullname, 1);1341 return;1342 }1343 if (S_ISDIR(entry.mode)) {1344 struct tree_desc sub;1345 struct pbase_tree_cache *tree;1346 const char *down = name+cmplen+1;1347 int downlen = name_cmp_len(down);13481349 tree = pbase_tree_get(entry.oid);1350 if (!tree)1351 return;1352 init_tree_desc(&sub, tree->tree_data, tree->tree_size);13531354 add_pbase_object(&sub, down, downlen, fullname);1355 pbase_tree_put(tree);1356 }1357 }1358}13591360static unsigned *done_pbase_paths;1361static int done_pbase_paths_num;1362static int done_pbase_paths_alloc;1363static int done_pbase_path_pos(unsigned hash)1364{1365 int lo = 0;1366 int hi = done_pbase_paths_num;1367 while (lo < hi) {1368 int mi = lo + (hi - lo) / 2;1369 if (done_pbase_paths[mi] == hash)1370 return mi;1371 if (done_pbase_paths[mi] < hash)1372 hi = mi;1373 else1374 lo = mi + 1;1375 }1376 return -lo-1;1377}13781379static int check_pbase_path(unsigned hash)1380{1381 int pos = done_pbase_path_pos(hash);1382 if (0 <= pos)1383 return 1;1384 pos = -pos - 1;1385 ALLOC_GROW(done_pbase_paths,1386 done_pbase_paths_num + 1,1387 done_pbase_paths_alloc);1388 done_pbase_paths_num++;1389 if (pos < done_pbase_paths_num)1390 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,1391 done_pbase_paths_num - pos - 1);1392 done_pbase_paths[pos] = hash;1393 return 0;1394}13951396static void add_preferred_base_object(const char *name)1397{1398 struct pbase_tree *it;1399 int cmplen;1400 unsigned hash = pack_name_hash(name);14011402 if (!num_preferred_base || check_pbase_path(hash))1403 return;14041405 cmplen = name_cmp_len(name);1406 for (it = pbase_tree; it; it = it->next) {1407 if (cmplen == 0) {1408 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);1409 }1410 else {1411 struct tree_desc tree;1412 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1413 add_pbase_object(&tree, name, cmplen, name);1414 }1415 }1416}14171418static void add_preferred_base(struct object_id *oid)1419{1420 struct pbase_tree *it;1421 void *data;1422 unsigned long size;1423 struct object_id tree_oid;14241425 if (window <= num_preferred_base++)1426 return;14271428 data = read_object_with_reference(oid, tree_type, &size, &tree_oid);1429 if (!data)1430 return;14311432 for (it = pbase_tree; it; it = it->next) {1433 if (!oidcmp(&it->pcache.oid, &tree_oid)) {1434 free(data);1435 return;1436 }1437 }14381439 it = xcalloc(1, sizeof(*it));1440 it->next = pbase_tree;1441 pbase_tree = it;14421443 oidcpy(&it->pcache.oid, &tree_oid);1444 it->pcache.tree_data = data;1445 it->pcache.tree_size = size;1446}14471448static void cleanup_preferred_base(void)1449{1450 struct pbase_tree *it;1451 unsigned i;14521453 it = pbase_tree;1454 pbase_tree = NULL;1455 while (it) {1456 struct pbase_tree *tmp = it;1457 it = tmp->next;1458 free(tmp->pcache.tree_data);1459 free(tmp);1460 }14611462 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {1463 if (!pbase_tree_cache[i])1464 continue;1465 free(pbase_tree_cache[i]->tree_data);1466 FREE_AND_NULL(pbase_tree_cache[i]);1467 }14681469 FREE_AND_NULL(done_pbase_paths);1470 done_pbase_paths_num = done_pbase_paths_alloc = 0;1471}14721473static void check_object(struct object_entry *entry)1474{1475 unsigned long canonical_size;14761477 if (IN_PACK(entry)) {1478 struct packed_git *p = IN_PACK(entry);1479 struct pack_window *w_curs = NULL;1480 const unsigned char *base_ref = NULL;1481 struct object_entry *base_entry;1482 unsigned long used, used_0;1483 unsigned long avail;1484 off_t ofs;1485 unsigned char *buf, c;1486 enum object_type type;1487 unsigned long in_pack_size;14881489 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);14901491 /*1492 * We want in_pack_type even if we do not reuse delta1493 * since non-delta representations could still be reused.1494 */1495 used = unpack_object_header_buffer(buf, avail,1496 &type,1497 &in_pack_size);1498 if (used == 0)1499 goto give_up;15001501 if (type < 0)1502 BUG("invalid type %d", type);1503 entry->in_pack_type = type;15041505 /*1506 * Determine if this is a delta and if so whether we can1507 * reuse it or not. Otherwise let's find out as cheaply as1508 * possible what the actual type and size for this object is.1509 */1510 switch (entry->in_pack_type) {1511 default:1512 /* Not a delta hence we've already got all we need. */1513 oe_set_type(entry, entry->in_pack_type);1514 SET_SIZE(entry, in_pack_size);1515 entry->in_pack_header_size = used;1516 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)1517 goto give_up;1518 unuse_pack(&w_curs);1519 return;1520 case OBJ_REF_DELTA:1521 if (reuse_delta && !entry->preferred_base)1522 base_ref = use_pack(p, &w_curs,1523 entry->in_pack_offset + used, NULL);1524 entry->in_pack_header_size = used + the_hash_algo->rawsz;1525 break;1526 case OBJ_OFS_DELTA:1527 buf = use_pack(p, &w_curs,1528 entry->in_pack_offset + used, NULL);1529 used_0 = 0;1530 c = buf[used_0++];1531 ofs = c & 127;1532 while (c & 128) {1533 ofs += 1;1534 if (!ofs || MSB(ofs, 7)) {1535 error(_("delta base offset overflow in pack for %s"),1536 oid_to_hex(&entry->idx.oid));1537 goto give_up;1538 }1539 c = buf[used_0++];1540 ofs = (ofs << 7) + (c & 127);1541 }1542 ofs = entry->in_pack_offset - ofs;1543 if (ofs <= 0 || ofs >= entry->in_pack_offset) {1544 error(_("delta base offset out of bound for %s"),1545 oid_to_hex(&entry->idx.oid));1546 goto give_up;1547 }1548 if (reuse_delta && !entry->preferred_base) {1549 struct revindex_entry *revidx;1550 revidx = find_pack_revindex(p, ofs);1551 if (!revidx)1552 goto give_up;1553 base_ref = nth_packed_object_sha1(p, revidx->nr);1554 }1555 entry->in_pack_header_size = used + used_0;1556 break;1557 }15581559 if (base_ref && (1560 (base_entry = packlist_find(&to_pack, base_ref, NULL)) ||1561 (thin &&1562 bitmap_has_sha1_in_uninteresting(bitmap_git, base_ref))) &&1563 in_same_island(&entry->idx.oid, &base_entry->idx.oid)) {1564 /*1565 * If base_ref was set above that means we wish to1566 * reuse delta data, and either we found that object in1567 * the list of objects we want to pack, or it's one we1568 * know the receiver has.1569 *1570 * Depth value does not matter - find_deltas() will1571 * never consider reused delta as the base object to1572 * deltify other objects against, in order to avoid1573 * circular deltas.1574 */1575 oe_set_type(entry, entry->in_pack_type);1576 SET_SIZE(entry, in_pack_size); /* delta size */1577 SET_DELTA_SIZE(entry, in_pack_size);15781579 if (base_entry) {1580 SET_DELTA(entry, base_entry);1581 entry->delta_sibling_idx = base_entry->delta_child_idx;1582 SET_DELTA_CHILD(base_entry, entry);1583 } else {1584 SET_DELTA_EXT(entry, base_ref);1585 }15861587 unuse_pack(&w_curs);1588 return;1589 }15901591 if (oe_type(entry)) {1592 off_t delta_pos;15931594 /*1595 * This must be a delta and we already know what the1596 * final object type is. Let's extract the actual1597 * object size from the delta header.1598 */1599 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;1600 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);1601 if (canonical_size == 0)1602 goto give_up;1603 SET_SIZE(entry, canonical_size);1604 unuse_pack(&w_curs);1605 return;1606 }16071608 /*1609 * No choice but to fall back to the recursive delta walk1610 * with sha1_object_info() to find about the object type1611 * at this point...1612 */1613 give_up:1614 unuse_pack(&w_curs);1615 }16161617 oe_set_type(entry,1618 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1619 if (entry->type_valid) {1620 SET_SIZE(entry, canonical_size);1621 } else {1622 /*1623 * Bad object type is checked in prepare_pack(). This is1624 * to permit a missing preferred base object to be ignored1625 * as a preferred base. Doing so can result in a larger1626 * pack file, but the transfer will still take place.1627 */1628 }1629}16301631static int pack_offset_sort(const void *_a, const void *_b)1632{1633 const struct object_entry *a = *(struct object_entry **)_a;1634 const struct object_entry *b = *(struct object_entry **)_b;1635 const struct packed_git *a_in_pack = IN_PACK(a);1636 const struct packed_git *b_in_pack = IN_PACK(b);16371638 /* avoid filesystem trashing with loose objects */1639 if (!a_in_pack && !b_in_pack)1640 return oidcmp(&a->idx.oid, &b->idx.oid);16411642 if (a_in_pack < b_in_pack)1643 return -1;1644 if (a_in_pack > b_in_pack)1645 return 1;1646 return a->in_pack_offset < b->in_pack_offset ? -1 :1647 (a->in_pack_offset > b->in_pack_offset);1648}16491650/*1651 * Drop an on-disk delta we were planning to reuse. Naively, this would1652 * just involve blanking out the "delta" field, but we have to deal1653 * with some extra book-keeping:1654 *1655 * 1. Removing ourselves from the delta_sibling linked list.1656 *1657 * 2. Updating our size/type to the non-delta representation. These were1658 * either not recorded initially (size) or overwritten with the delta type1659 * (type) when check_object() decided to reuse the delta.1660 *1661 * 3. Resetting our delta depth, as we are now a base object.1662 */1663static void drop_reused_delta(struct object_entry *entry)1664{1665 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;1666 struct object_info oi = OBJECT_INFO_INIT;1667 enum object_type type;1668 unsigned long size;16691670 while (*idx) {1671 struct object_entry *oe = &to_pack.objects[*idx - 1];16721673 if (oe == entry)1674 *idx = oe->delta_sibling_idx;1675 else1676 idx = &oe->delta_sibling_idx;1677 }1678 SET_DELTA(entry, NULL);1679 entry->depth = 0;16801681 oi.sizep = &size;1682 oi.typep = &type;1683 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {1684 /*1685 * We failed to get the info from this pack for some reason;1686 * fall back to sha1_object_info, which may find another copy.1687 * And if that fails, the error will be recorded in oe_type(entry)1688 * and dealt with in prepare_pack().1689 */1690 oe_set_type(entry,1691 oid_object_info(the_repository, &entry->idx.oid, &size));1692 } else {1693 oe_set_type(entry, type);1694 }1695 SET_SIZE(entry, size);1696}16971698/*1699 * Follow the chain of deltas from this entry onward, throwing away any links1700 * that cause us to hit a cycle (as determined by the DFS state flags in1701 * the entries).1702 *1703 * We also detect too-long reused chains that would violate our --depth1704 * limit.1705 */1706static void break_delta_chains(struct object_entry *entry)1707{1708 /*1709 * The actual depth of each object we will write is stored as an int,1710 * as it cannot exceed our int "depth" limit. But before we break1711 * changes based no that limit, we may potentially go as deep as the1712 * number of objects, which is elsewhere bounded to a uint32_t.1713 */1714 uint32_t total_depth;1715 struct object_entry *cur, *next;17161717 for (cur = entry, total_depth = 0;1718 cur;1719 cur = DELTA(cur), total_depth++) {1720 if (cur->dfs_state == DFS_DONE) {1721 /*1722 * We've already seen this object and know it isn't1723 * part of a cycle. We do need to append its depth1724 * to our count.1725 */1726 total_depth += cur->depth;1727 break;1728 }17291730 /*1731 * We break cycles before looping, so an ACTIVE state (or any1732 * other cruft which made its way into the state variable)1733 * is a bug.1734 */1735 if (cur->dfs_state != DFS_NONE)1736 BUG("confusing delta dfs state in first pass: %d",1737 cur->dfs_state);17381739 /*1740 * Now we know this is the first time we've seen the object. If1741 * it's not a delta, we're done traversing, but we'll mark it1742 * done to save time on future traversals.1743 */1744 if (!DELTA(cur)) {1745 cur->dfs_state = DFS_DONE;1746 break;1747 }17481749 /*1750 * Mark ourselves as active and see if the next step causes1751 * us to cycle to another active object. It's important to do1752 * this _before_ we loop, because it impacts where we make the1753 * cut, and thus how our total_depth counter works.1754 * E.g., We may see a partial loop like:1755 *1756 * A -> B -> C -> D -> B1757 *1758 * Cutting B->C breaks the cycle. But now the depth of A is1759 * only 1, and our total_depth counter is at 3. The size of the1760 * error is always one less than the size of the cycle we1761 * broke. Commits C and D were "lost" from A's chain.1762 *1763 * If we instead cut D->B, then the depth of A is correct at 3.1764 * We keep all commits in the chain that we examined.1765 */1766 cur->dfs_state = DFS_ACTIVE;1767 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {1768 drop_reused_delta(cur);1769 cur->dfs_state = DFS_DONE;1770 break;1771 }1772 }17731774 /*1775 * And now that we've gone all the way to the bottom of the chain, we1776 * need to clear the active flags and set the depth fields as1777 * appropriate. Unlike the loop above, which can quit when it drops a1778 * delta, we need to keep going to look for more depth cuts. So we need1779 * an extra "next" pointer to keep going after we reset cur->delta.1780 */1781 for (cur = entry; cur; cur = next) {1782 next = DELTA(cur);17831784 /*1785 * We should have a chain of zero or more ACTIVE states down to1786 * a final DONE. We can quit after the DONE, because either it1787 * has no bases, or we've already handled them in a previous1788 * call.1789 */1790 if (cur->dfs_state == DFS_DONE)1791 break;1792 else if (cur->dfs_state != DFS_ACTIVE)1793 BUG("confusing delta dfs state in second pass: %d",1794 cur->dfs_state);17951796 /*1797 * If the total_depth is more than depth, then we need to snip1798 * the chain into two or more smaller chains that don't exceed1799 * the maximum depth. Most of the resulting chains will contain1800 * (depth + 1) entries (i.e., depth deltas plus one base), and1801 * the last chain (i.e., the one containing entry) will contain1802 * whatever entries are left over, namely1803 * (total_depth % (depth + 1)) of them.1804 *1805 * Since we are iterating towards decreasing depth, we need to1806 * decrement total_depth as we go, and we need to write to the1807 * entry what its final depth will be after all of the1808 * snipping. Since we're snipping into chains of length (depth1809 * + 1) entries, the final depth of an entry will be its1810 * original depth modulo (depth + 1). Any time we encounter an1811 * entry whose final depth is supposed to be zero, we snip it1812 * from its delta base, thereby making it so.1813 */1814 cur->depth = (total_depth--) % (depth + 1);1815 if (!cur->depth)1816 drop_reused_delta(cur);18171818 cur->dfs_state = DFS_DONE;1819 }1820}18211822static void get_object_details(void)1823{1824 uint32_t i;1825 struct object_entry **sorted_by_offset;18261827 if (progress)1828 progress_state = start_progress(_("Counting objects"),1829 to_pack.nr_objects);18301831 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));1832 for (i = 0; i < to_pack.nr_objects; i++)1833 sorted_by_offset[i] = to_pack.objects + i;1834 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);18351836 for (i = 0; i < to_pack.nr_objects; i++) {1837 struct object_entry *entry = sorted_by_offset[i];1838 check_object(entry);1839 if (entry->type_valid &&1840 oe_size_greater_than(&to_pack, entry, big_file_threshold))1841 entry->no_try_delta = 1;1842 display_progress(progress_state, i + 1);1843 }1844 stop_progress(&progress_state);18451846 /*1847 * This must happen in a second pass, since we rely on the delta1848 * information for the whole list being completed.1849 */1850 for (i = 0; i < to_pack.nr_objects; i++)1851 break_delta_chains(&to_pack.objects[i]);18521853 free(sorted_by_offset);1854}18551856/*1857 * We search for deltas in a list sorted by type, by filename hash, and then1858 * by size, so that we see progressively smaller and smaller files.1859 * That's because we prefer deltas to be from the bigger file1860 * to the smaller -- deletes are potentially cheaper, but perhaps1861 * more importantly, the bigger file is likely the more recent1862 * one. The deepest deltas are therefore the oldest objects which are1863 * less susceptible to be accessed often.1864 */1865static int type_size_sort(const void *_a, const void *_b)1866{1867 const struct object_entry *a = *(struct object_entry **)_a;1868 const struct object_entry *b = *(struct object_entry **)_b;1869 enum object_type a_type = oe_type(a);1870 enum object_type b_type = oe_type(b);1871 unsigned long a_size = SIZE(a);1872 unsigned long b_size = SIZE(b);18731874 if (a_type > b_type)1875 return -1;1876 if (a_type < b_type)1877 return 1;1878 if (a->hash > b->hash)1879 return -1;1880 if (a->hash < b->hash)1881 return 1;1882 if (a->preferred_base > b->preferred_base)1883 return -1;1884 if (a->preferred_base < b->preferred_base)1885 return 1;1886 if (use_delta_islands) {1887 int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);1888 if (island_cmp)1889 return island_cmp;1890 }1891 if (a_size > b_size)1892 return -1;1893 if (a_size < b_size)1894 return 1;1895 return a < b ? -1 : (a > b); /* newest first */1896}18971898struct unpacked {1899 struct object_entry *entry;1900 void *data;1901 struct delta_index *index;1902 unsigned depth;1903};19041905static int delta_cacheable(unsigned long src_size, unsigned long trg_size,1906 unsigned long delta_size)1907{1908 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1909 return 0;19101911 if (delta_size < cache_max_small_delta_size)1912 return 1;19131914 /* cache delta, if objects are large enough compared to delta size */1915 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))1916 return 1;19171918 return 0;1919}19201921#ifndef NO_PTHREADS19221923/* Protect access to object database */1924static pthread_mutex_t read_mutex;1925#define read_lock() pthread_mutex_lock(&read_mutex)1926#define read_unlock() pthread_mutex_unlock(&read_mutex)19271928/* Protect delta_cache_size */1929static pthread_mutex_t cache_mutex;1930#define cache_lock() pthread_mutex_lock(&cache_mutex)1931#define cache_unlock() pthread_mutex_unlock(&cache_mutex)19321933/*1934 * Protect object list partitioning (e.g. struct thread_param) and1935 * progress_state1936 */1937static pthread_mutex_t progress_mutex;1938#define progress_lock() pthread_mutex_lock(&progress_mutex)1939#define progress_unlock() pthread_mutex_unlock(&progress_mutex)19401941/*1942 * Access to struct object_entry is unprotected since each thread owns1943 * a portion of the main object list. Just don't access object entries1944 * ahead in the list because they can be stolen and would need1945 * progress_mutex for protection.1946 */1947#else19481949#define read_lock() (void)01950#define read_unlock() (void)01951#define cache_lock() (void)01952#define cache_unlock() (void)01953#define progress_lock() (void)01954#define progress_unlock() (void)019551956#endif19571958/*1959 * Return the size of the object without doing any delta1960 * reconstruction (so non-deltas are true object sizes, but deltas1961 * return the size of the delta data).1962 */1963unsigned long oe_get_size_slow(struct packing_data *pack,1964 const struct object_entry *e)1965{1966 struct packed_git *p;1967 struct pack_window *w_curs;1968 unsigned char *buf;1969 enum object_type type;1970 unsigned long used, avail, size;19711972 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {1973 read_lock();1974 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)1975 die(_("unable to get size of %s"),1976 oid_to_hex(&e->idx.oid));1977 read_unlock();1978 return size;1979 }19801981 p = oe_in_pack(pack, e);1982 if (!p)1983 BUG("when e->type is a delta, it must belong to a pack");19841985 read_lock();1986 w_curs = NULL;1987 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);1988 used = unpack_object_header_buffer(buf, avail, &type, &size);1989 if (used == 0)1990 die(_("unable to parse object header of %s"),1991 oid_to_hex(&e->idx.oid));19921993 unuse_pack(&w_curs);1994 read_unlock();1995 return size;1996}19971998static int try_delta(struct unpacked *trg, struct unpacked *src,1999 unsigned max_depth, unsigned long *mem_usage)2000{2001 struct object_entry *trg_entry = trg->entry;2002 struct object_entry *src_entry = src->entry;2003 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;2004 unsigned ref_depth;2005 enum object_type type;2006 void *delta_buf;20072008 /* Don't bother doing diffs between different types */2009 if (oe_type(trg_entry) != oe_type(src_entry))2010 return -1;20112012 /*2013 * We do not bother to try a delta that we discarded on an2014 * earlier try, but only when reusing delta data. Note that2015 * src_entry that is marked as the preferred_base should always2016 * be considered, as even if we produce a suboptimal delta against2017 * it, we will still save the transfer cost, as we already know2018 * the other side has it and we won't send src_entry at all.2019 */2020 if (reuse_delta && IN_PACK(trg_entry) &&2021 IN_PACK(trg_entry) == IN_PACK(src_entry) &&2022 !src_entry->preferred_base &&2023 trg_entry->in_pack_type != OBJ_REF_DELTA &&2024 trg_entry->in_pack_type != OBJ_OFS_DELTA)2025 return 0;20262027 /* Let's not bust the allowed depth. */2028 if (src->depth >= max_depth)2029 return 0;20302031 /* Now some size filtering heuristics. */2032 trg_size = SIZE(trg_entry);2033 if (!DELTA(trg_entry)) {2034 max_size = trg_size/2 - the_hash_algo->rawsz;2035 ref_depth = 1;2036 } else {2037 max_size = DELTA_SIZE(trg_entry);2038 ref_depth = trg->depth;2039 }2040 max_size = (uint64_t)max_size * (max_depth - src->depth) /2041 (max_depth - ref_depth + 1);2042 if (max_size == 0)2043 return 0;2044 src_size = SIZE(src_entry);2045 sizediff = src_size < trg_size ? trg_size - src_size : 0;2046 if (sizediff >= max_size)2047 return 0;2048 if (trg_size < src_size / 32)2049 return 0;20502051 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))2052 return 0;20532054 /* Load data if not already done */2055 if (!trg->data) {2056 read_lock();2057 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);2058 read_unlock();2059 if (!trg->data)2060 die(_("object %s cannot be read"),2061 oid_to_hex(&trg_entry->idx.oid));2062 if (sz != trg_size)2063 die(_("object %s inconsistent object length (%lu vs %lu)"),2064 oid_to_hex(&trg_entry->idx.oid), sz,2065 trg_size);2066 *mem_usage += sz;2067 }2068 if (!src->data) {2069 read_lock();2070 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);2071 read_unlock();2072 if (!src->data) {2073 if (src_entry->preferred_base) {2074 static int warned = 0;2075 if (!warned++)2076 warning(_("object %s cannot be read"),2077 oid_to_hex(&src_entry->idx.oid));2078 /*2079 * Those objects are not included in the2080 * resulting pack. Be resilient and ignore2081 * them if they can't be read, in case the2082 * pack could be created nevertheless.2083 */2084 return 0;2085 }2086 die(_("object %s cannot be read"),2087 oid_to_hex(&src_entry->idx.oid));2088 }2089 if (sz != src_size)2090 die(_("object %s inconsistent object length (%lu vs %lu)"),2091 oid_to_hex(&src_entry->idx.oid), sz,2092 src_size);2093 *mem_usage += sz;2094 }2095 if (!src->index) {2096 src->index = create_delta_index(src->data, src_size);2097 if (!src->index) {2098 static int warned = 0;2099 if (!warned++)2100 warning(_("suboptimal pack - out of memory"));2101 return 0;2102 }2103 *mem_usage += sizeof_delta_index(src->index);2104 }21052106 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2107 if (!delta_buf)2108 return 0;21092110 if (DELTA(trg_entry)) {2111 /* Prefer only shallower same-sized deltas. */2112 if (delta_size == DELTA_SIZE(trg_entry) &&2113 src->depth + 1 >= trg->depth) {2114 free(delta_buf);2115 return 0;2116 }2117 }21182119 /*2120 * Handle memory allocation outside of the cache2121 * accounting lock. Compiler will optimize the strangeness2122 * away when NO_PTHREADS is defined.2123 */2124 free(trg_entry->delta_data);2125 cache_lock();2126 if (trg_entry->delta_data) {2127 delta_cache_size -= DELTA_SIZE(trg_entry);2128 trg_entry->delta_data = NULL;2129 }2130 if (delta_cacheable(src_size, trg_size, delta_size)) {2131 delta_cache_size += delta_size;2132 cache_unlock();2133 trg_entry->delta_data = xrealloc(delta_buf, delta_size);2134 } else {2135 cache_unlock();2136 free(delta_buf);2137 }21382139 SET_DELTA(trg_entry, src_entry);2140 SET_DELTA_SIZE(trg_entry, delta_size);2141 trg->depth = src->depth + 1;21422143 return 1;2144}21452146static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)2147{2148 struct object_entry *child = DELTA_CHILD(me);2149 unsigned int m = n;2150 while (child) {2151 unsigned int c = check_delta_limit(child, n + 1);2152 if (m < c)2153 m = c;2154 child = DELTA_SIBLING(child);2155 }2156 return m;2157}21582159static unsigned long free_unpacked(struct unpacked *n)2160{2161 unsigned long freed_mem = sizeof_delta_index(n->index);2162 free_delta_index(n->index);2163 n->index = NULL;2164 if (n->data) {2165 freed_mem += SIZE(n->entry);2166 FREE_AND_NULL(n->data);2167 }2168 n->entry = NULL;2169 n->depth = 0;2170 return freed_mem;2171}21722173static void find_deltas(struct object_entry **list, unsigned *list_size,2174 int window, int depth, unsigned *processed)2175{2176 uint32_t i, idx = 0, count = 0;2177 struct unpacked *array;2178 unsigned long mem_usage = 0;21792180 array = xcalloc(window, sizeof(struct unpacked));21812182 for (;;) {2183 struct object_entry *entry;2184 struct unpacked *n = array + idx;2185 int j, max_depth, best_base = -1;21862187 progress_lock();2188 if (!*list_size) {2189 progress_unlock();2190 break;2191 }2192 entry = *list++;2193 (*list_size)--;2194 if (!entry->preferred_base) {2195 (*processed)++;2196 display_progress(progress_state, *processed);2197 }2198 progress_unlock();21992200 mem_usage -= free_unpacked(n);2201 n->entry = entry;22022203 while (window_memory_limit &&2204 mem_usage > window_memory_limit &&2205 count > 1) {2206 uint32_t tail = (idx + window - count) % window;2207 mem_usage -= free_unpacked(array + tail);2208 count--;2209 }22102211 /* We do not compute delta to *create* objects we are not2212 * going to pack.2213 */2214 if (entry->preferred_base)2215 goto next;22162217 /*2218 * If the current object is at pack edge, take the depth the2219 * objects that depend on the current object into account2220 * otherwise they would become too deep.2221 */2222 max_depth = depth;2223 if (DELTA_CHILD(entry)) {2224 max_depth -= check_delta_limit(entry, 0);2225 if (max_depth <= 0)2226 goto next;2227 }22282229 j = window;2230 while (--j > 0) {2231 int ret;2232 uint32_t other_idx = idx + j;2233 struct unpacked *m;2234 if (other_idx >= window)2235 other_idx -= window;2236 m = array + other_idx;2237 if (!m->entry)2238 break;2239 ret = try_delta(n, m, max_depth, &mem_usage);2240 if (ret < 0)2241 break;2242 else if (ret > 0)2243 best_base = other_idx;2244 }22452246 /*2247 * If we decided to cache the delta data, then it is best2248 * to compress it right away. First because we have to do2249 * it anyway, and doing it here while we're threaded will2250 * save a lot of time in the non threaded write phase,2251 * as well as allow for caching more deltas within2252 * the same cache size limit.2253 * ...2254 * But only if not writing to stdout, since in that case2255 * the network is most likely throttling writes anyway,2256 * and therefore it is best to go to the write phase ASAP2257 * instead, as we can afford spending more time compressing2258 * between writes at that moment.2259 */2260 if (entry->delta_data && !pack_to_stdout) {2261 unsigned long size;22622263 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));2264 if (size < (1U << OE_Z_DELTA_BITS)) {2265 entry->z_delta_size = size;2266 cache_lock();2267 delta_cache_size -= DELTA_SIZE(entry);2268 delta_cache_size += entry->z_delta_size;2269 cache_unlock();2270 } else {2271 FREE_AND_NULL(entry->delta_data);2272 entry->z_delta_size = 0;2273 }2274 }22752276 /* if we made n a delta, and if n is already at max2277 * depth, leaving it in the window is pointless. we2278 * should evict it first.2279 */2280 if (DELTA(entry) && max_depth <= n->depth)2281 continue;22822283 /*2284 * Move the best delta base up in the window, after the2285 * currently deltified object, to keep it longer. It will2286 * be the first base object to be attempted next.2287 */2288 if (DELTA(entry)) {2289 struct unpacked swap = array[best_base];2290 int dist = (window + idx - best_base) % window;2291 int dst = best_base;2292 while (dist--) {2293 int src = (dst + 1) % window;2294 array[dst] = array[src];2295 dst = src;2296 }2297 array[dst] = swap;2298 }22992300 next:2301 idx++;2302 if (count + 1 < window)2303 count++;2304 if (idx >= window)2305 idx = 0;2306 }23072308 for (i = 0; i < window; ++i) {2309 free_delta_index(array[i].index);2310 free(array[i].data);2311 }2312 free(array);2313}23142315#ifndef NO_PTHREADS23162317static void try_to_free_from_threads(size_t size)2318{2319 read_lock();2320 release_pack_memory(size);2321 read_unlock();2322}23232324static try_to_free_t old_try_to_free_routine;23252326/*2327 * The main object list is split into smaller lists, each is handed to2328 * one worker.2329 *2330 * The main thread waits on the condition that (at least) one of the workers2331 * has stopped working (which is indicated in the .working member of2332 * struct thread_params).2333 *2334 * When a work thread has completed its work, it sets .working to 0 and2335 * signals the main thread and waits on the condition that .data_ready2336 * becomes 1.2337 *2338 * The main thread steals half of the work from the worker that has2339 * most work left to hand it to the idle worker.2340 */23412342struct thread_params {2343 pthread_t thread;2344 struct object_entry **list;2345 unsigned list_size;2346 unsigned remaining;2347 int window;2348 int depth;2349 int working;2350 int data_ready;2351 pthread_mutex_t mutex;2352 pthread_cond_t cond;2353 unsigned *processed;2354};23552356static pthread_cond_t progress_cond;23572358/*2359 * Mutex and conditional variable can't be statically-initialized on Windows.2360 */2361static void init_threaded_search(void)2362{2363 init_recursive_mutex(&read_mutex);2364 pthread_mutex_init(&cache_mutex, NULL);2365 pthread_mutex_init(&progress_mutex, NULL);2366 pthread_cond_init(&progress_cond, NULL);2367 pthread_mutex_init(&to_pack.lock, NULL);2368 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);2369}23702371static void cleanup_threaded_search(void)2372{2373 set_try_to_free_routine(old_try_to_free_routine);2374 pthread_cond_destroy(&progress_cond);2375 pthread_mutex_destroy(&read_mutex);2376 pthread_mutex_destroy(&cache_mutex);2377 pthread_mutex_destroy(&progress_mutex);2378}23792380static void *threaded_find_deltas(void *arg)2381{2382 struct thread_params *me = arg;23832384 progress_lock();2385 while (me->remaining) {2386 progress_unlock();23872388 find_deltas(me->list, &me->remaining,2389 me->window, me->depth, me->processed);23902391 progress_lock();2392 me->working = 0;2393 pthread_cond_signal(&progress_cond);2394 progress_unlock();23952396 /*2397 * We must not set ->data_ready before we wait on the2398 * condition because the main thread may have set it to 12399 * before we get here. In order to be sure that new2400 * work is available if we see 1 in ->data_ready, it2401 * was initialized to 0 before this thread was spawned2402 * and we reset it to 0 right away.2403 */2404 pthread_mutex_lock(&me->mutex);2405 while (!me->data_ready)2406 pthread_cond_wait(&me->cond, &me->mutex);2407 me->data_ready = 0;2408 pthread_mutex_unlock(&me->mutex);24092410 progress_lock();2411 }2412 progress_unlock();2413 /* leave ->working 1 so that this doesn't get more work assigned */2414 return NULL;2415}24162417static void ll_find_deltas(struct object_entry **list, unsigned list_size,2418 int window, int depth, unsigned *processed)2419{2420 struct thread_params *p;2421 int i, ret, active_threads = 0;24222423 init_threaded_search();24242425 if (delta_search_threads <= 1) {2426 find_deltas(list, &list_size, window, depth, processed);2427 cleanup_threaded_search();2428 return;2429 }2430 if (progress > pack_to_stdout)2431 fprintf_ln(stderr, _("Delta compression using up to %d threads"),2432 delta_search_threads);2433 p = xcalloc(delta_search_threads, sizeof(*p));24342435 /* Partition the work amongst work threads. */2436 for (i = 0; i < delta_search_threads; i++) {2437 unsigned sub_size = list_size / (delta_search_threads - i);24382439 /* don't use too small segments or no deltas will be found */2440 if (sub_size < 2*window && i+1 < delta_search_threads)2441 sub_size = 0;24422443 p[i].window = window;2444 p[i].depth = depth;2445 p[i].processed = processed;2446 p[i].working = 1;2447 p[i].data_ready = 0;24482449 /* try to split chunks on "path" boundaries */2450 while (sub_size && sub_size < list_size &&2451 list[sub_size]->hash &&2452 list[sub_size]->hash == list[sub_size-1]->hash)2453 sub_size++;24542455 p[i].list = list;2456 p[i].list_size = sub_size;2457 p[i].remaining = sub_size;24582459 list += sub_size;2460 list_size -= sub_size;2461 }24622463 /* Start work threads. */2464 for (i = 0; i < delta_search_threads; i++) {2465 if (!p[i].list_size)2466 continue;2467 pthread_mutex_init(&p[i].mutex, NULL);2468 pthread_cond_init(&p[i].cond, NULL);2469 ret = pthread_create(&p[i].thread, NULL,2470 threaded_find_deltas, &p[i]);2471 if (ret)2472 die(_("unable to create thread: %s"), strerror(ret));2473 active_threads++;2474 }24752476 /*2477 * Now let's wait for work completion. Each time a thread is done2478 * with its work, we steal half of the remaining work from the2479 * thread with the largest number of unprocessed objects and give2480 * it to that newly idle thread. This ensure good load balancing2481 * until the remaining object list segments are simply too short2482 * to be worth splitting anymore.2483 */2484 while (active_threads) {2485 struct thread_params *target = NULL;2486 struct thread_params *victim = NULL;2487 unsigned sub_size = 0;24882489 progress_lock();2490 for (;;) {2491 for (i = 0; !target && i < delta_search_threads; i++)2492 if (!p[i].working)2493 target = &p[i];2494 if (target)2495 break;2496 pthread_cond_wait(&progress_cond, &progress_mutex);2497 }24982499 for (i = 0; i < delta_search_threads; i++)2500 if (p[i].remaining > 2*window &&2501 (!victim || victim->remaining < p[i].remaining))2502 victim = &p[i];2503 if (victim) {2504 sub_size = victim->remaining / 2;2505 list = victim->list + victim->list_size - sub_size;2506 while (sub_size && list[0]->hash &&2507 list[0]->hash == list[-1]->hash) {2508 list++;2509 sub_size--;2510 }2511 if (!sub_size) {2512 /*2513 * It is possible for some "paths" to have2514 * so many objects that no hash boundary2515 * might be found. Let's just steal the2516 * exact half in that case.2517 */2518 sub_size = victim->remaining / 2;2519 list -= sub_size;2520 }2521 target->list = list;2522 victim->list_size -= sub_size;2523 victim->remaining -= sub_size;2524 }2525 target->list_size = sub_size;2526 target->remaining = sub_size;2527 target->working = 1;2528 progress_unlock();25292530 pthread_mutex_lock(&target->mutex);2531 target->data_ready = 1;2532 pthread_cond_signal(&target->cond);2533 pthread_mutex_unlock(&target->mutex);25342535 if (!sub_size) {2536 pthread_join(target->thread, NULL);2537 pthread_cond_destroy(&target->cond);2538 pthread_mutex_destroy(&target->mutex);2539 active_threads--;2540 }2541 }2542 cleanup_threaded_search();2543 free(p);2544}25452546#else2547#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)2548#endif25492550static void add_tag_chain(const struct object_id *oid)2551{2552 struct tag *tag;25532554 /*2555 * We catch duplicates already in add_object_entry(), but we'd2556 * prefer to do this extra check to avoid having to parse the2557 * tag at all if we already know that it's being packed (e.g., if2558 * it was included via bitmaps, we would not have parsed it2559 * previously).2560 */2561 if (packlist_find(&to_pack, oid->hash, NULL))2562 return;25632564 tag = lookup_tag(the_repository, oid);2565 while (1) {2566 if (!tag || parse_tag(tag) || !tag->tagged)2567 die(_("unable to pack objects reachable from tag %s"),2568 oid_to_hex(oid));25692570 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);25712572 if (tag->tagged->type != OBJ_TAG)2573 return;25742575 tag = (struct tag *)tag->tagged;2576 }2577}25782579static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)2580{2581 struct object_id peeled;25822583 if (starts_with(path, "refs/tags/") && /* is a tag? */2584 !peel_ref(path, &peeled) && /* peelable? */2585 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */2586 add_tag_chain(oid);2587 return 0;2588}25892590static void prepare_pack(int window, int depth)2591{2592 struct object_entry **delta_list;2593 uint32_t i, nr_deltas;2594 unsigned n;25952596 if (use_delta_islands)2597 resolve_tree_islands(progress, &to_pack);25982599 get_object_details();26002601 /*2602 * If we're locally repacking then we need to be doubly careful2603 * from now on in order to make sure no stealth corruption gets2604 * propagated to the new pack. Clients receiving streamed packs2605 * should validate everything they get anyway so no need to incur2606 * the additional cost here in that case.2607 */2608 if (!pack_to_stdout)2609 do_check_packed_object_crc = 1;26102611 if (!to_pack.nr_objects || !window || !depth)2612 return;26132614 ALLOC_ARRAY(delta_list, to_pack.nr_objects);2615 nr_deltas = n = 0;26162617 for (i = 0; i < to_pack.nr_objects; i++) {2618 struct object_entry *entry = to_pack.objects + i;26192620 if (DELTA(entry))2621 /* This happens if we decided to reuse existing2622 * delta from a pack. "reuse_delta &&" is implied.2623 */2624 continue;26252626 if (!entry->type_valid ||2627 oe_size_less_than(&to_pack, entry, 50))2628 continue;26292630 if (entry->no_try_delta)2631 continue;26322633 if (!entry->preferred_base) {2634 nr_deltas++;2635 if (oe_type(entry) < 0)2636 die(_("unable to get type of object %s"),2637 oid_to_hex(&entry->idx.oid));2638 } else {2639 if (oe_type(entry) < 0) {2640 /*2641 * This object is not found, but we2642 * don't have to include it anyway.2643 */2644 continue;2645 }2646 }26472648 delta_list[n++] = entry;2649 }26502651 if (nr_deltas && n > 1) {2652 unsigned nr_done = 0;2653 if (progress)2654 progress_state = start_progress(_("Compressing objects"),2655 nr_deltas);2656 QSORT(delta_list, n, type_size_sort);2657 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2658 stop_progress(&progress_state);2659 if (nr_done != nr_deltas)2660 die(_("inconsistency with delta count"));2661 }2662 free(delta_list);2663}26642665static int git_pack_config(const char *k, const char *v, void *cb)2666{2667 if (!strcmp(k, "pack.window")) {2668 window = git_config_int(k, v);2669 return 0;2670 }2671 if (!strcmp(k, "pack.windowmemory")) {2672 window_memory_limit = git_config_ulong(k, v);2673 return 0;2674 }2675 if (!strcmp(k, "pack.depth")) {2676 depth = git_config_int(k, v);2677 return 0;2678 }2679 if (!strcmp(k, "pack.deltacachesize")) {2680 max_delta_cache_size = git_config_int(k, v);2681 return 0;2682 }2683 if (!strcmp(k, "pack.deltacachelimit")) {2684 cache_max_small_delta_size = git_config_int(k, v);2685 return 0;2686 }2687 if (!strcmp(k, "pack.writebitmaphashcache")) {2688 if (git_config_bool(k, v))2689 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2690 else2691 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2692 }2693 if (!strcmp(k, "pack.usebitmaps")) {2694 use_bitmap_index_default = git_config_bool(k, v);2695 return 0;2696 }2697 if (!strcmp(k, "pack.threads")) {2698 delta_search_threads = git_config_int(k, v);2699 if (delta_search_threads < 0)2700 die(_("invalid number of threads specified (%d)"),2701 delta_search_threads);2702#ifdef NO_PTHREADS2703 if (delta_search_threads != 1) {2704 warning(_("no threads support, ignoring %s"), k);2705 delta_search_threads = 0;2706 }2707#endif2708 return 0;2709 }2710 if (!strcmp(k, "pack.indexversion")) {2711 pack_idx_opts.version = git_config_int(k, v);2712 if (pack_idx_opts.version > 2)2713 die(_("bad pack.indexversion=%"PRIu32),2714 pack_idx_opts.version);2715 return 0;2716 }2717 return git_default_config(k, v, cb);2718}27192720static void read_object_list_from_stdin(void)2721{2722 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];2723 struct object_id oid;2724 const char *p;27252726 for (;;) {2727 if (!fgets(line, sizeof(line), stdin)) {2728 if (feof(stdin))2729 break;2730 if (!ferror(stdin))2731 die("BUG: fgets returned NULL, not EOF, not error!");2732 if (errno != EINTR)2733 die_errno("fgets");2734 clearerr(stdin);2735 continue;2736 }2737 if (line[0] == '-') {2738 if (get_oid_hex(line+1, &oid))2739 die(_("expected edge object ID, got garbage:\n %s"),2740 line);2741 add_preferred_base(&oid);2742 continue;2743 }2744 if (parse_oid_hex(line, &oid, &p))2745 die(_("expected object ID, got garbage:\n %s"), line);27462747 add_preferred_base_object(p + 1);2748 add_object_entry(&oid, OBJ_NONE, p + 1, 0);2749 }2750}27512752/* Remember to update object flag allocation in object.h */2753#define OBJECT_ADDED (1u<<20)27542755static void show_commit(struct commit *commit, void *data)2756{2757 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);2758 commit->object.flags |= OBJECT_ADDED;27592760 if (write_bitmap_index)2761 index_commit_for_bitmap(commit);27622763 if (use_delta_islands)2764 propagate_island_marks(commit);2765}27662767static void show_object(struct object *obj, const char *name, void *data)2768{2769 add_preferred_base_object(name);2770 add_object_entry(&obj->oid, obj->type, name, 0);2771 obj->flags |= OBJECT_ADDED;27722773 if (use_delta_islands) {2774 const char *p;2775 unsigned depth = 0;2776 struct object_entry *ent;27772778 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))2779 depth++;27802781 ent = packlist_find(&to_pack, obj->oid.hash, NULL);2782 if (ent && depth > oe_tree_depth(&to_pack, ent))2783 oe_set_tree_depth(&to_pack, ent, depth);2784 }2785}27862787static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)2788{2789 assert(arg_missing_action == MA_ALLOW_ANY);27902791 /*2792 * Quietly ignore ALL missing objects. This avoids problems with2793 * staging them now and getting an odd error later.2794 */2795 if (!has_object_file(&obj->oid))2796 return;27972798 show_object(obj, name, data);2799}28002801static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)2802{2803 assert(arg_missing_action == MA_ALLOW_PROMISOR);28042805 /*2806 * Quietly ignore EXPECTED missing objects. This avoids problems with2807 * staging them now and getting an odd error later.2808 */2809 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))2810 return;28112812 show_object(obj, name, data);2813}28142815static int option_parse_missing_action(const struct option *opt,2816 const char *arg, int unset)2817{2818 assert(arg);2819 assert(!unset);28202821 if (!strcmp(arg, "error")) {2822 arg_missing_action = MA_ERROR;2823 fn_show_object = show_object;2824 return 0;2825 }28262827 if (!strcmp(arg, "allow-any")) {2828 arg_missing_action = MA_ALLOW_ANY;2829 fetch_if_missing = 0;2830 fn_show_object = show_object__ma_allow_any;2831 return 0;2832 }28332834 if (!strcmp(arg, "allow-promisor")) {2835 arg_missing_action = MA_ALLOW_PROMISOR;2836 fetch_if_missing = 0;2837 fn_show_object = show_object__ma_allow_promisor;2838 return 0;2839 }28402841 die(_("invalid value for --missing"));2842 return 0;2843}28442845static void show_edge(struct commit *commit)2846{2847 add_preferred_base(&commit->object.oid);2848}28492850struct in_pack_object {2851 off_t offset;2852 struct object *object;2853};28542855struct in_pack {2856 unsigned int alloc;2857 unsigned int nr;2858 struct in_pack_object *array;2859};28602861static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)2862{2863 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);2864 in_pack->array[in_pack->nr].object = object;2865 in_pack->nr++;2866}28672868/*2869 * Compare the objects in the offset order, in order to emulate the2870 * "git rev-list --objects" output that produced the pack originally.2871 */2872static int ofscmp(const void *a_, const void *b_)2873{2874 struct in_pack_object *a = (struct in_pack_object *)a_;2875 struct in_pack_object *b = (struct in_pack_object *)b_;28762877 if (a->offset < b->offset)2878 return -1;2879 else if (a->offset > b->offset)2880 return 1;2881 else2882 return oidcmp(&a->object->oid, &b->object->oid);2883}28842885static void add_objects_in_unpacked_packs(struct rev_info *revs)2886{2887 struct packed_git *p;2888 struct in_pack in_pack;2889 uint32_t i;28902891 memset(&in_pack, 0, sizeof(in_pack));28922893 for (p = get_all_packs(the_repository); p; p = p->next) {2894 struct object_id oid;2895 struct object *o;28962897 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)2898 continue;2899 if (open_pack_index(p))2900 die(_("cannot open pack index"));29012902 ALLOC_GROW(in_pack.array,2903 in_pack.nr + p->num_objects,2904 in_pack.alloc);29052906 for (i = 0; i < p->num_objects; i++) {2907 nth_packed_object_oid(&oid, p, i);2908 o = lookup_unknown_object(oid.hash);2909 if (!(o->flags & OBJECT_ADDED))2910 mark_in_pack_object(o, p, &in_pack);2911 o->flags |= OBJECT_ADDED;2912 }2913 }29142915 if (in_pack.nr) {2916 QSORT(in_pack.array, in_pack.nr, ofscmp);2917 for (i = 0; i < in_pack.nr; i++) {2918 struct object *o = in_pack.array[i].object;2919 add_object_entry(&o->oid, o->type, "", 0);2920 }2921 }2922 free(in_pack.array);2923}29242925static int add_loose_object(const struct object_id *oid, const char *path,2926 void *data)2927{2928 enum object_type type = oid_object_info(the_repository, oid, NULL);29292930 if (type < 0) {2931 warning(_("loose object at %s could not be examined"), path);2932 return 0;2933 }29342935 add_object_entry(oid, type, "", 0);2936 return 0;2937}29382939/*2940 * We actually don't even have to worry about reachability here.2941 * add_object_entry will weed out duplicates, so we just add every2942 * loose object we find.2943 */2944static void add_unreachable_loose_objects(void)2945{2946 for_each_loose_file_in_objdir(get_object_directory(),2947 add_loose_object,2948 NULL, NULL, NULL);2949}29502951static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2952{2953 static struct packed_git *last_found = (void *)1;2954 struct packed_git *p;29552956 p = (last_found != (void *)1) ? last_found :2957 get_all_packs(the_repository);29582959 while (p) {2960 if ((!p->pack_local || p->pack_keep ||2961 p->pack_keep_in_core) &&2962 find_pack_entry_one(oid->hash, p)) {2963 last_found = p;2964 return 1;2965 }2966 if (p == last_found)2967 p = get_all_packs(the_repository);2968 else2969 p = p->next;2970 if (p == last_found)2971 p = p->next;2972 }2973 return 0;2974}29752976/*2977 * Store a list of sha1s that are should not be discarded2978 * because they are either written too recently, or are2979 * reachable from another object that was.2980 *2981 * This is filled by get_object_list.2982 */2983static struct oid_array recent_objects;29842985static int loosened_object_can_be_discarded(const struct object_id *oid,2986 timestamp_t mtime)2987{2988 if (!unpack_unreachable_expiration)2989 return 0;2990 if (mtime > unpack_unreachable_expiration)2991 return 0;2992 if (oid_array_lookup(&recent_objects, oid) >= 0)2993 return 0;2994 return 1;2995}29962997static void loosen_unused_packed_objects(struct rev_info *revs)2998{2999 struct packed_git *p;3000 uint32_t i;3001 struct object_id oid;30023003 for (p = get_all_packs(the_repository); p; p = p->next) {3004 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)3005 continue;30063007 if (open_pack_index(p))3008 die(_("cannot open pack index"));30093010 for (i = 0; i < p->num_objects; i++) {3011 nth_packed_object_oid(&oid, p, i);3012 if (!packlist_find(&to_pack, oid.hash, NULL) &&3013 !has_sha1_pack_kept_or_nonlocal(&oid) &&3014 !loosened_object_can_be_discarded(&oid, p->mtime))3015 if (force_object_loose(&oid, p->mtime))3016 die(_("unable to force loose object"));3017 }3018 }3019}30203021/*3022 * This tracks any options which pack-reuse code expects to be on, or which a3023 * reader of the pack might not understand, and which would therefore prevent3024 * blind reuse of what we have on disk.3025 */3026static int pack_options_allow_reuse(void)3027{3028 return pack_to_stdout &&3029 allow_ofs_delta &&3030 !ignore_packed_keep_on_disk &&3031 !ignore_packed_keep_in_core &&3032 (!local || !have_non_local_packs) &&3033 !incremental;3034}30353036static int get_object_list_from_bitmap(struct rev_info *revs)3037{3038 if (!(bitmap_git = prepare_bitmap_walk(revs)))3039 return -1;30403041 if (pack_options_allow_reuse() &&3042 !reuse_partial_packfile_from_bitmap(3043 bitmap_git,3044 &reuse_packfile,3045 &reuse_packfile_objects,3046 &reuse_packfile_offset)) {3047 assert(reuse_packfile_objects);3048 nr_result += reuse_packfile_objects;3049 display_progress(progress_state, nr_result);3050 }30513052 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);3053 return 0;3054}30553056static void record_recent_object(struct object *obj,3057 const char *name,3058 void *data)3059{3060 oid_array_append(&recent_objects, &obj->oid);3061}30623063static void record_recent_commit(struct commit *commit, void *data)3064{3065 oid_array_append(&recent_objects, &commit->object.oid);3066}30673068static void get_object_list(int ac, const char **av)3069{3070 struct rev_info revs;3071 char line[1000];3072 int flags = 0;30733074 init_revisions(&revs, NULL);3075 save_commit_buffer = 0;3076 setup_revisions(ac, av, &revs, NULL);30773078 /* make sure shallows are read */3079 is_repository_shallow(the_repository);30803081 while (fgets(line, sizeof(line), stdin) != NULL) {3082 int len = strlen(line);3083 if (len && line[len - 1] == '\n')3084 line[--len] = 0;3085 if (!len)3086 break;3087 if (*line == '-') {3088 if (!strcmp(line, "--not")) {3089 flags ^= UNINTERESTING;3090 write_bitmap_index = 0;3091 continue;3092 }3093 if (starts_with(line, "--shallow ")) {3094 struct object_id oid;3095 if (get_oid_hex(line + 10, &oid))3096 die("not an SHA-1 '%s'", line + 10);3097 register_shallow(the_repository, &oid);3098 use_bitmap_index = 0;3099 continue;3100 }3101 die(_("not a rev '%s'"), line);3102 }3103 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3104 die(_("bad revision '%s'"), line);3105 }31063107 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))3108 return;31093110 if (use_delta_islands)3111 load_delta_islands();31123113 if (prepare_revision_walk(&revs))3114 die(_("revision walk setup failed"));3115 mark_edges_uninteresting(&revs, show_edge);31163117 if (!fn_show_object)3118 fn_show_object = show_object;3119 traverse_commit_list_filtered(&filter_options, &revs,3120 show_commit, fn_show_object, NULL,3121 NULL);31223123 if (unpack_unreachable_expiration) {3124 revs.ignore_missing_links = 1;3125 if (add_unseen_recent_objects_to_traversal(&revs,3126 unpack_unreachable_expiration))3127 die(_("unable to add recent objects"));3128 if (prepare_revision_walk(&revs))3129 die(_("revision walk setup failed"));3130 traverse_commit_list(&revs, record_recent_commit,3131 record_recent_object, NULL);3132 }31333134 if (keep_unreachable)3135 add_objects_in_unpacked_packs(&revs);3136 if (pack_loose_unreachable)3137 add_unreachable_loose_objects();3138 if (unpack_unreachable)3139 loosen_unused_packed_objects(&revs);31403141 oid_array_clear(&recent_objects);3142}31433144static void add_extra_kept_packs(const struct string_list *names)3145{3146 struct packed_git *p;31473148 if (!names->nr)3149 return;31503151 for (p = get_all_packs(the_repository); p; p = p->next) {3152 const char *name = basename(p->pack_name);3153 int i;31543155 if (!p->pack_local)3156 continue;31573158 for (i = 0; i < names->nr; i++)3159 if (!fspathcmp(name, names->items[i].string))3160 break;31613162 if (i < names->nr) {3163 p->pack_keep_in_core = 1;3164 ignore_packed_keep_in_core = 1;3165 continue;3166 }3167 }3168}31693170static int option_parse_index_version(const struct option *opt,3171 const char *arg, int unset)3172{3173 char *c;3174 const char *val = arg;3175 pack_idx_opts.version = strtoul(val, &c, 10);3176 if (pack_idx_opts.version > 2)3177 die(_("unsupported index version %s"), val);3178 if (*c == ',' && c[1])3179 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);3180 if (*c || pack_idx_opts.off32_limit & 0x80000000)3181 die(_("bad index version '%s'"), val);3182 return 0;3183}31843185static int option_parse_unpack_unreachable(const struct option *opt,3186 const char *arg, int unset)3187{3188 if (unset) {3189 unpack_unreachable = 0;3190 unpack_unreachable_expiration = 0;3191 }3192 else {3193 unpack_unreachable = 1;3194 if (arg)3195 unpack_unreachable_expiration = approxidate(arg);3196 }3197 return 0;3198}31993200int cmd_pack_objects(int argc, const char **argv, const char *prefix)3201{3202 int use_internal_rev_list = 0;3203 int shallow = 0;3204 int all_progress_implied = 0;3205 struct argv_array rp = ARGV_ARRAY_INIT;3206 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;3207 int rev_list_index = 0;3208 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3209 struct option pack_objects_options[] = {3210 OPT_SET_INT('q', "quiet", &progress,3211 N_("do not show progress meter"), 0),3212 OPT_SET_INT(0, "progress", &progress,3213 N_("show progress meter"), 1),3214 OPT_SET_INT(0, "all-progress", &progress,3215 N_("show progress meter during object writing phase"), 2),3216 OPT_BOOL(0, "all-progress-implied",3217 &all_progress_implied,3218 N_("similar to --all-progress when progress meter is shown")),3219 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),3220 N_("write the pack index file in the specified idx format version"),3221 0, option_parse_index_version },3222 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,3223 N_("maximum size of each output pack file")),3224 OPT_BOOL(0, "local", &local,3225 N_("ignore borrowed objects from alternate object store")),3226 OPT_BOOL(0, "incremental", &incremental,3227 N_("ignore packed objects")),3228 OPT_INTEGER(0, "window", &window,3229 N_("limit pack window by objects")),3230 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,3231 N_("limit pack window by memory in addition to object limit")),3232 OPT_INTEGER(0, "depth", &depth,3233 N_("maximum length of delta chain allowed in the resulting pack")),3234 OPT_BOOL(0, "reuse-delta", &reuse_delta,3235 N_("reuse existing deltas")),3236 OPT_BOOL(0, "reuse-object", &reuse_object,3237 N_("reuse existing objects")),3238 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,3239 N_("use OFS_DELTA objects")),3240 OPT_INTEGER(0, "threads", &delta_search_threads,3241 N_("use threads when searching for best delta matches")),3242 OPT_BOOL(0, "non-empty", &non_empty,3243 N_("do not create an empty pack output")),3244 OPT_BOOL(0, "revs", &use_internal_rev_list,3245 N_("read revision arguments from standard input")),3246 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,3247 N_("limit the objects to those that are not yet packed"),3248 1, PARSE_OPT_NONEG),3249 OPT_SET_INT_F(0, "all", &rev_list_all,3250 N_("include objects reachable from any reference"),3251 1, PARSE_OPT_NONEG),3252 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,3253 N_("include objects referred by reflog entries"),3254 1, PARSE_OPT_NONEG),3255 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,3256 N_("include objects referred to by the index"),3257 1, PARSE_OPT_NONEG),3258 OPT_BOOL(0, "stdout", &pack_to_stdout,3259 N_("output pack to stdout")),3260 OPT_BOOL(0, "include-tag", &include_tag,3261 N_("include tag objects that refer to objects to be packed")),3262 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,3263 N_("keep unreachable objects")),3264 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,3265 N_("pack loose unreachable objects")),3266 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),3267 N_("unpack unreachable objects newer than <time>"),3268 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3269 OPT_BOOL(0, "thin", &thin,3270 N_("create thin packs")),3271 OPT_BOOL(0, "shallow", &shallow,3272 N_("create packs suitable for shallow fetches")),3273 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,3274 N_("ignore packs that have companion .keep file")),3275 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),3276 N_("ignore this pack")),3277 OPT_INTEGER(0, "compression", &pack_compression_level,3278 N_("pack compression level")),3279 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,3280 N_("do not hide commits by grafts"), 0),3281 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,3282 N_("use a bitmap index if available to speed up counting objects")),3283 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,3284 N_("write a bitmap index together with the pack index")),3285 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3286 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),3287 N_("handling for missing objects"), PARSE_OPT_NONEG,3288 option_parse_missing_action },3289 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,3290 N_("do not pack objects in promisor packfiles")),3291 OPT_BOOL(0, "delta-islands", &use_delta_islands,3292 N_("respect islands during delta compression")),3293 OPT_END(),3294 };32953296 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))3297 BUG("too many dfs states, increase OE_DFS_STATE_BITS");32983299 read_replace_refs = 0;33003301 reset_pack_idx_option(&pack_idx_opts);3302 git_config(git_pack_config, NULL);33033304 progress = isatty(2);3305 argc = parse_options(argc, argv, prefix, pack_objects_options,3306 pack_usage, 0);33073308 if (argc) {3309 base_name = argv[0];3310 argc--;3311 }3312 if (pack_to_stdout != !base_name || argc)3313 usage_with_options(pack_usage, pack_objects_options);33143315 if (depth >= (1 << OE_DEPTH_BITS)) {3316 warning(_("delta chain depth %d is too deep, forcing %d"),3317 depth, (1 << OE_DEPTH_BITS) - 1);3318 depth = (1 << OE_DEPTH_BITS) - 1;3319 }3320 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {3321 warning(_("pack.deltaCacheLimit is too high, forcing %d"),3322 (1U << OE_Z_DELTA_BITS) - 1);3323 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;3324 }33253326 argv_array_push(&rp, "pack-objects");3327 if (thin) {3328 use_internal_rev_list = 1;3329 argv_array_push(&rp, shallow3330 ? "--objects-edge-aggressive"3331 : "--objects-edge");3332 } else3333 argv_array_push(&rp, "--objects");33343335 if (rev_list_all) {3336 use_internal_rev_list = 1;3337 argv_array_push(&rp, "--all");3338 }3339 if (rev_list_reflog) {3340 use_internal_rev_list = 1;3341 argv_array_push(&rp, "--reflog");3342 }3343 if (rev_list_index) {3344 use_internal_rev_list = 1;3345 argv_array_push(&rp, "--indexed-objects");3346 }3347 if (rev_list_unpacked) {3348 use_internal_rev_list = 1;3349 argv_array_push(&rp, "--unpacked");3350 }33513352 if (exclude_promisor_objects) {3353 use_internal_rev_list = 1;3354 fetch_if_missing = 0;3355 argv_array_push(&rp, "--exclude-promisor-objects");3356 }3357 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)3358 use_internal_rev_list = 1;33593360 if (!reuse_object)3361 reuse_delta = 0;3362 if (pack_compression_level == -1)3363 pack_compression_level = Z_DEFAULT_COMPRESSION;3364 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)3365 die(_("bad pack compression level %d"), pack_compression_level);33663367 if (!delta_search_threads) /* --threads=0 means autodetect */3368 delta_search_threads = online_cpus();33693370#ifdef NO_PTHREADS3371 if (delta_search_threads != 1)3372 warning(_("no threads support, ignoring --threads"));3373#endif3374 if (!pack_to_stdout && !pack_size_limit)3375 pack_size_limit = pack_size_limit_cfg;3376 if (pack_to_stdout && pack_size_limit)3377 die(_("--max-pack-size cannot be used to build a pack for transfer"));3378 if (pack_size_limit && pack_size_limit < 1024*1024) {3379 warning(_("minimum pack size limit is 1 MiB"));3380 pack_size_limit = 1024*1024;3381 }33823383 if (!pack_to_stdout && thin)3384 die(_("--thin cannot be used to build an indexable pack"));33853386 if (keep_unreachable && unpack_unreachable)3387 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));3388 if (!rev_list_all || !rev_list_reflog || !rev_list_index)3389 unpack_unreachable_expiration = 0;33903391 if (filter_options.choice) {3392 if (!pack_to_stdout)3393 die(_("cannot use --filter without --stdout"));3394 use_bitmap_index = 0;3395 }33963397 /*3398 * "soft" reasons not to use bitmaps - for on-disk repack by default we want3399 *3400 * - to produce good pack (with bitmap index not-yet-packed objects are3401 * packed in suboptimal order).3402 *3403 * - to use more robust pack-generation codepath (avoiding possible3404 * bugs in bitmap code and possible bitmap index corruption).3405 */3406 if (!pack_to_stdout)3407 use_bitmap_index_default = 0;34083409 if (use_bitmap_index < 0)3410 use_bitmap_index = use_bitmap_index_default;34113412 /* "hard" reasons not to use bitmaps; these just won't work at all */3413 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))3414 use_bitmap_index = 0;34153416 if (pack_to_stdout || !rev_list_all)3417 write_bitmap_index = 0;34183419 if (use_delta_islands)3420 argv_array_push(&rp, "--topo-order");34213422 if (progress && all_progress_implied)3423 progress = 2;34243425 add_extra_kept_packs(&keep_pack_list);3426 if (ignore_packed_keep_on_disk) {3427 struct packed_git *p;3428 for (p = get_all_packs(the_repository); p; p = p->next)3429 if (p->pack_local && p->pack_keep)3430 break;3431 if (!p) /* no keep-able packs found */3432 ignore_packed_keep_on_disk = 0;3433 }3434 if (local) {3435 /*3436 * unlike ignore_packed_keep_on_disk above, we do not3437 * want to unset "local" based on looking at packs, as3438 * it also covers non-local objects3439 */3440 struct packed_git *p;3441 for (p = get_all_packs(the_repository); p; p = p->next) {3442 if (!p->pack_local) {3443 have_non_local_packs = 1;3444 break;3445 }3446 }3447 }34483449 prepare_packing_data(&to_pack);34503451 if (progress)3452 progress_state = start_progress(_("Enumerating objects"), 0);3453 if (!use_internal_rev_list)3454 read_object_list_from_stdin();3455 else {3456 get_object_list(rp.argc, rp.argv);3457 argv_array_clear(&rp);3458 }3459 cleanup_preferred_base();3460 if (include_tag && nr_result)3461 for_each_ref(add_ref_tag, NULL);3462 stop_progress(&progress_state);34633464 if (non_empty && !nr_result)3465 return 0;3466 if (nr_result)3467 prepare_pack(window, depth);3468 write_pack_file();3469 if (progress)3470 fprintf_ln(stderr,3471 _("Total %"PRIu32" (delta %"PRIu32"),"3472 " reused %"PRIu32" (delta %"PRIu32")"),3473 written, written_delta, reused, reused_delta);3474 return 0;3475}