11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
16 #include "list-objects.h"
22 #include "thread-utils.h"
25 static const char pack_usage[] =
26 "git pack-objects [ -q | --progress | --all-progress ]\n"
27 " [--all-progress-implied]\n"
28 " [--max-pack-size=<n>] [--local] [--incremental]\n"
29 " [--window=<n>] [--window-memory=<n>] [--depth=<n>]\n"
30 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
31 " [--threads=<n>] [--non-empty] [--revs [--unpacked | --all]]\n"
32 " [--reflog] [--stdout | base-name] [--include-tag]\n"
33 " [--keep-unreachable | --unpack-unreachable]\n"
34 " [< ref-list | < object-list]";
37 struct pack_idx_entry idx;
38 unsigned long size; /* uncompressed size */
39 struct packed_git *in_pack; /* already in pack */
41 struct object_entry *delta; /* delta base object */
42 struct object_entry *delta_child; /* deltified objects who bases me */
43 struct object_entry *delta_sibling; /* other deltified objects who
44 * uses the same base as me
46 void *delta_data; /* cached delta (uncompressed) */
47 unsigned long delta_size; /* delta data size (uncompressed) */
48 unsigned long z_delta_size; /* delta data size (compressed) */
49 unsigned int hash; /* name hint hash */
50 enum object_type type;
51 enum object_type in_pack_type; /* could be delta */
52 unsigned char in_pack_header_size;
53 unsigned char preferred_base; /* we do not pack this, but is available
54 * to be used as the base object to delta
57 unsigned char no_try_delta;
61 * Objects we are going to pack are collected in objects array (dynamically
62 * expanded). nr_objects & nr_alloc controls this array. They are stored
63 * in the order we see -- typically rev-list --objects order that gives us
64 * nice "minimum seek" order.
66 static struct object_entry *objects;
67 static struct pack_idx_entry **written_list;
68 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
71 static int reuse_delta = 1, reuse_object = 1;
72 static int keep_unreachable, unpack_unreachable, include_tag;
74 static int incremental;
75 static int ignore_packed_keep;
76 static int allow_ofs_delta;
77 static const char *base_name;
78 static int progress = 1;
79 static int window = 10;
80 static unsigned long pack_size_limit, pack_size_limit_cfg;
81 static int depth = 50;
82 static int delta_search_threads;
83 static int pack_to_stdout;
84 static int num_preferred_base;
85 static struct progress *progress_state;
86 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
87 static int pack_compression_seen;
89 static unsigned long delta_cache_size = 0;
90 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
91 static unsigned long cache_max_small_delta_size = 1000;
93 static unsigned long window_memory_limit = 0;
96 * The object names in objects array are hashed with this hashtable,
97 * to help looking up the entry by object name.
98 * This hashtable is built after all the objects are seen.
100 static int *object_ix;
101 static int object_ix_hashsz;
106 static uint32_t written, written_delta;
107 static uint32_t reused, reused_delta;
110 static void *get_delta(struct object_entry *entry)
112 unsigned long size, base_size, delta_size;
113 void *buf, *base_buf, *delta_buf;
114 enum object_type type;
116 buf = read_sha1_file(entry->idx.sha1, &type, &size);
118 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
119 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
121 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
122 delta_buf = diff_delta(base_buf, base_size,
123 buf, size, &delta_size, 0);
124 if (!delta_buf || delta_size != entry->delta_size)
125 die("delta size changed");
131 static unsigned long do_compress(void **pptr, unsigned long size)
135 unsigned long maxsize;
137 memset(&stream, 0, sizeof(stream));
138 deflateInit(&stream, pack_compression_level);
139 maxsize = deflateBound(&stream, size);
142 out = xmalloc(maxsize);
146 stream.avail_in = size;
147 stream.next_out = out;
148 stream.avail_out = maxsize;
149 while (deflate(&stream, Z_FINISH) == Z_OK)
154 return stream.total_out;
158 * we are going to reuse the existing object data as is. make
159 * sure it is not corrupt.
161 static int check_pack_inflate(struct packed_git *p,
162 struct pack_window **w_curs,
165 unsigned long expect)
168 unsigned char fakebuf[4096], *in;
171 memset(&stream, 0, sizeof(stream));
172 git_inflate_init(&stream);
174 in = use_pack(p, w_curs, offset, &stream.avail_in);
176 stream.next_out = fakebuf;
177 stream.avail_out = sizeof(fakebuf);
178 st = git_inflate(&stream, Z_FINISH);
179 offset += stream.next_in - in;
180 } while (st == Z_OK || st == Z_BUF_ERROR);
181 git_inflate_end(&stream);
182 return (st == Z_STREAM_END &&
183 stream.total_out == expect &&
184 stream.total_in == len) ? 0 : -1;
187 static void copy_pack_data(struct sha1file *f,
188 struct packed_git *p,
189 struct pack_window **w_curs,
197 in = use_pack(p, w_curs, offset, &avail);
199 avail = (unsigned int)len;
200 sha1write(f, in, avail);
206 static unsigned long write_object(struct sha1file *f,
207 struct object_entry *entry,
210 unsigned long size, limit, datalen;
212 unsigned char header[10], dheader[10];
214 enum object_type type;
215 int usable_delta, to_reuse;
222 /* apply size limit if limited packsize and not first object */
223 if (!pack_size_limit || !nr_written)
225 else if (pack_size_limit <= write_offset)
227 * the earlier object did not fit the limit; avoid
228 * mistaking this with unlimited (i.e. limit = 0).
232 limit = pack_size_limit - write_offset;
235 usable_delta = 0; /* no delta */
236 else if (!pack_size_limit)
237 usable_delta = 1; /* unlimited packfile */
238 else if (entry->delta->idx.offset == (off_t)-1)
239 usable_delta = 0; /* base was written to another pack */
240 else if (entry->delta->idx.offset)
241 usable_delta = 1; /* base already exists in this pack */
243 usable_delta = 0; /* base could end up in another pack */
246 to_reuse = 0; /* explicit */
247 else if (!entry->in_pack)
248 to_reuse = 0; /* can't reuse what we don't have */
249 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
250 /* check_object() decided it for us ... */
251 to_reuse = usable_delta;
252 /* ... but pack split may override that */
253 else if (type != entry->in_pack_type)
254 to_reuse = 0; /* pack has delta which is unusable */
255 else if (entry->delta)
256 to_reuse = 0; /* we want to pack afresh */
258 to_reuse = 1; /* we have it in-pack undeltified,
259 * and we do not need to deltify it.
265 buf = read_sha1_file(entry->idx.sha1, &type, &size);
267 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
269 * make sure no cached delta data remains from a
270 * previous attempt before a pack split occurred.
272 free(entry->delta_data);
273 entry->delta_data = NULL;
274 entry->z_delta_size = 0;
275 } else if (entry->delta_data) {
276 size = entry->delta_size;
277 buf = entry->delta_data;
278 entry->delta_data = NULL;
279 type = (allow_ofs_delta && entry->delta->idx.offset) ?
280 OBJ_OFS_DELTA : OBJ_REF_DELTA;
282 buf = get_delta(entry);
283 size = entry->delta_size;
284 type = (allow_ofs_delta && entry->delta->idx.offset) ?
285 OBJ_OFS_DELTA : OBJ_REF_DELTA;
288 if (entry->z_delta_size)
289 datalen = entry->z_delta_size;
291 datalen = do_compress(&buf, size);
294 * The object header is a byte of 'type' followed by zero or
295 * more bytes of length.
297 hdrlen = encode_in_pack_object_header(type, size, header);
299 if (type == OBJ_OFS_DELTA) {
301 * Deltas with relative base contain an additional
302 * encoding of the relative offset for the delta
303 * base from this object's position in the pack.
305 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
306 unsigned pos = sizeof(dheader) - 1;
307 dheader[pos] = ofs & 127;
309 dheader[--pos] = 128 | (--ofs & 127);
310 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
314 sha1write(f, header, hdrlen);
315 sha1write(f, dheader + pos, sizeof(dheader) - pos);
316 hdrlen += sizeof(dheader) - pos;
317 } else if (type == OBJ_REF_DELTA) {
319 * Deltas with a base reference contain
320 * an additional 20 bytes for the base sha1.
322 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
326 sha1write(f, header, hdrlen);
327 sha1write(f, entry->delta->idx.sha1, 20);
330 if (limit && hdrlen + datalen + 20 >= limit) {
334 sha1write(f, header, hdrlen);
336 sha1write(f, buf, datalen);
340 struct packed_git *p = entry->in_pack;
341 struct pack_window *w_curs = NULL;
342 struct revindex_entry *revidx;
346 type = (allow_ofs_delta && entry->delta->idx.offset) ?
347 OBJ_OFS_DELTA : OBJ_REF_DELTA;
348 hdrlen = encode_in_pack_object_header(type, entry->size, header);
350 offset = entry->in_pack_offset;
351 revidx = find_pack_revindex(p, offset);
352 datalen = revidx[1].offset - offset;
353 if (!pack_to_stdout && p->index_version > 1 &&
354 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
355 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
360 offset += entry->in_pack_header_size;
361 datalen -= entry->in_pack_header_size;
362 if (!pack_to_stdout && p->index_version == 1 &&
363 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
364 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
369 if (type == OBJ_OFS_DELTA) {
370 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
371 unsigned pos = sizeof(dheader) - 1;
372 dheader[pos] = ofs & 127;
374 dheader[--pos] = 128 | (--ofs & 127);
375 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
379 sha1write(f, header, hdrlen);
380 sha1write(f, dheader + pos, sizeof(dheader) - pos);
381 hdrlen += sizeof(dheader) - pos;
383 } else if (type == OBJ_REF_DELTA) {
384 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
388 sha1write(f, header, hdrlen);
389 sha1write(f, entry->delta->idx.sha1, 20);
393 if (limit && hdrlen + datalen + 20 >= limit) {
397 sha1write(f, header, hdrlen);
399 copy_pack_data(f, p, &w_curs, offset, datalen);
407 entry->idx.crc32 = crc32_end(f);
408 return hdrlen + datalen;
411 static int write_one(struct sha1file *f,
412 struct object_entry *e,
417 /* offset is non zero if object is written already. */
418 if (e->idx.offset || e->preferred_base)
421 /* if we are deltified, write out base object first. */
422 if (e->delta && !write_one(f, e->delta, offset))
425 e->idx.offset = *offset;
426 size = write_object(f, e, *offset);
431 written_list[nr_written++] = &e->idx;
433 /* make sure off_t is sufficiently large not to wrap */
434 if (signed_add_overflows(*offset, size))
435 die("pack too large for current definition of off_t");
440 static void write_pack_file(void)
445 struct pack_header hdr;
446 uint32_t nr_remaining = nr_result;
447 time_t last_mtime = 0;
449 if (progress > pack_to_stdout)
450 progress_state = start_progress("Writing objects", nr_result);
451 written_list = xmalloc(nr_objects * sizeof(*written_list));
454 unsigned char sha1[20];
455 char *pack_tmp_name = NULL;
457 if (pack_to_stdout) {
458 f = sha1fd_throughput(1, "<stdout>", progress_state);
460 char tmpname[PATH_MAX];
462 fd = odb_mkstemp(tmpname, sizeof(tmpname),
463 "pack/tmp_pack_XXXXXX");
464 pack_tmp_name = xstrdup(tmpname);
465 f = sha1fd(fd, pack_tmp_name);
468 hdr.hdr_signature = htonl(PACK_SIGNATURE);
469 hdr.hdr_version = htonl(PACK_VERSION);
470 hdr.hdr_entries = htonl(nr_remaining);
471 sha1write(f, &hdr, sizeof(hdr));
472 offset = sizeof(hdr);
474 for (; i < nr_objects; i++) {
475 if (!write_one(f, objects + i, &offset))
477 display_progress(progress_state, written);
481 * Did we write the wrong # entries in the header?
482 * If so, rewrite it like in fast-import
484 if (pack_to_stdout) {
485 sha1close(f, sha1, CSUM_CLOSE);
486 } else if (nr_written == nr_remaining) {
487 sha1close(f, sha1, CSUM_FSYNC);
489 int fd = sha1close(f, sha1, 0);
490 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
491 nr_written, sha1, offset);
495 if (!pack_to_stdout) {
497 const char *idx_tmp_name;
498 char tmpname[PATH_MAX];
500 idx_tmp_name = write_idx_file(NULL, written_list,
503 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
504 base_name, sha1_to_hex(sha1));
505 free_pack_by_name(tmpname);
506 if (adjust_shared_perm(pack_tmp_name))
507 die_errno("unable to make temporary pack file readable");
508 if (rename(pack_tmp_name, tmpname))
509 die_errno("unable to rename temporary pack file");
512 * Packs are runtime accessed in their mtime
513 * order since newer packs are more likely to contain
514 * younger objects. So if we are creating multiple
515 * packs then we should modify the mtime of later ones
516 * to preserve this property.
518 if (stat(tmpname, &st) < 0) {
519 warning("failed to stat %s: %s",
520 tmpname, strerror(errno));
521 } else if (!last_mtime) {
522 last_mtime = st.st_mtime;
525 utb.actime = st.st_atime;
526 utb.modtime = --last_mtime;
527 if (utime(tmpname, &utb) < 0)
528 warning("failed utime() on %s: %s",
529 tmpname, strerror(errno));
532 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
533 base_name, sha1_to_hex(sha1));
534 if (adjust_shared_perm(idx_tmp_name))
535 die_errno("unable to make temporary index file readable");
536 if (rename(idx_tmp_name, tmpname))
537 die_errno("unable to rename temporary index file");
539 free((void *) idx_tmp_name);
541 puts(sha1_to_hex(sha1));
544 /* mark written objects as written to previous pack */
545 for (j = 0; j < nr_written; j++) {
546 written_list[j]->offset = (off_t)-1;
548 nr_remaining -= nr_written;
549 } while (nr_remaining && i < nr_objects);
552 stop_progress(&progress_state);
553 if (written != nr_result)
554 die("wrote %"PRIu32" objects while expecting %"PRIu32,
558 static int locate_object_entry_hash(const unsigned char *sha1)
562 memcpy(&ui, sha1, sizeof(unsigned int));
563 i = ui % object_ix_hashsz;
564 while (0 < object_ix[i]) {
565 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
567 if (++i == object_ix_hashsz)
573 static struct object_entry *locate_object_entry(const unsigned char *sha1)
577 if (!object_ix_hashsz)
580 i = locate_object_entry_hash(sha1);
582 return &objects[object_ix[i]-1];
586 static void rehash_objects(void)
589 struct object_entry *oe;
591 object_ix_hashsz = nr_objects * 3;
592 if (object_ix_hashsz < 1024)
593 object_ix_hashsz = 1024;
594 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
595 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
596 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
597 int ix = locate_object_entry_hash(oe->idx.sha1);
601 object_ix[ix] = i + 1;
605 static unsigned name_hash(const char *name)
607 unsigned c, hash = 0;
613 * This effectively just creates a sortable number from the
614 * last sixteen non-whitespace characters. Last characters
615 * count "most", so things that end in ".c" sort together.
617 while ((c = *name++) != 0) {
620 hash = (hash >> 2) + (c << 24);
625 static void setup_delta_attr_check(struct git_attr_check *check)
627 static struct git_attr *attr_delta;
630 attr_delta = git_attr("delta");
632 check[0].attr = attr_delta;
635 static int no_try_delta(const char *path)
637 struct git_attr_check check[1];
639 setup_delta_attr_check(check);
640 if (git_checkattr(path, ARRAY_SIZE(check), check))
642 if (ATTR_FALSE(check->value))
647 static int add_object_entry(const unsigned char *sha1, enum object_type type,
648 const char *name, int exclude)
650 struct object_entry *entry;
651 struct packed_git *p, *found_pack = NULL;
652 off_t found_offset = 0;
654 unsigned hash = name_hash(name);
656 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
659 entry = objects + object_ix[ix] - 1;
660 if (!entry->preferred_base)
662 entry->preferred_base = 1;
667 if (!exclude && local && has_loose_object_nonlocal(sha1))
670 for (p = packed_git; p; p = p->next) {
671 off_t offset = find_pack_entry_one(sha1, p);
674 found_offset = offset;
681 if (local && !p->pack_local)
683 if (ignore_packed_keep && p->pack_local && p->pack_keep)
688 if (nr_objects >= nr_alloc) {
689 nr_alloc = (nr_alloc + 1024) * 3 / 2;
690 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
693 entry = objects + nr_objects++;
694 memset(entry, 0, sizeof(*entry));
695 hashcpy(entry->idx.sha1, sha1);
700 entry->preferred_base = 1;
704 entry->in_pack = found_pack;
705 entry->in_pack_offset = found_offset;
708 if (object_ix_hashsz * 3 <= nr_objects * 4)
711 object_ix[-1 - ix] = nr_objects;
713 display_progress(progress_state, nr_objects);
715 if (name && no_try_delta(name))
716 entry->no_try_delta = 1;
721 struct pbase_tree_cache {
722 unsigned char sha1[20];
726 unsigned long tree_size;
729 static struct pbase_tree_cache *(pbase_tree_cache[256]);
730 static int pbase_tree_cache_ix(const unsigned char *sha1)
732 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
734 static int pbase_tree_cache_ix_incr(int ix)
736 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
739 static struct pbase_tree {
740 struct pbase_tree *next;
741 /* This is a phony "cache" entry; we are not
742 * going to evict it nor find it through _get()
743 * mechanism -- this is for the toplevel node that
744 * would almost always change with any commit.
746 struct pbase_tree_cache pcache;
749 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
751 struct pbase_tree_cache *ent, *nent;
754 enum object_type type;
756 int my_ix = pbase_tree_cache_ix(sha1);
757 int available_ix = -1;
759 /* pbase-tree-cache acts as a limited hashtable.
760 * your object will be found at your index or within a few
761 * slots after that slot if it is cached.
763 for (neigh = 0; neigh < 8; neigh++) {
764 ent = pbase_tree_cache[my_ix];
765 if (ent && !hashcmp(ent->sha1, sha1)) {
769 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
770 ((0 <= available_ix) &&
771 (!ent && pbase_tree_cache[available_ix])))
772 available_ix = my_ix;
775 my_ix = pbase_tree_cache_ix_incr(my_ix);
778 /* Did not find one. Either we got a bogus request or
779 * we need to read and perhaps cache.
781 data = read_sha1_file(sha1, &type, &size);
784 if (type != OBJ_TREE) {
789 /* We need to either cache or return a throwaway copy */
791 if (available_ix < 0)
794 ent = pbase_tree_cache[available_ix];
795 my_ix = available_ix;
799 nent = xmalloc(sizeof(*nent));
800 nent->temporary = (available_ix < 0);
803 /* evict and reuse */
804 free(ent->tree_data);
807 hashcpy(nent->sha1, sha1);
808 nent->tree_data = data;
809 nent->tree_size = size;
811 if (!nent->temporary)
812 pbase_tree_cache[my_ix] = nent;
816 static void pbase_tree_put(struct pbase_tree_cache *cache)
818 if (!cache->temporary) {
822 free(cache->tree_data);
826 static int name_cmp_len(const char *name)
829 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
834 static void add_pbase_object(struct tree_desc *tree,
837 const char *fullname)
839 struct name_entry entry;
842 while (tree_entry(tree,&entry)) {
843 if (S_ISGITLINK(entry.mode))
845 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
846 memcmp(name, entry.path, cmplen);
851 if (name[cmplen] != '/') {
852 add_object_entry(entry.sha1,
853 object_type(entry.mode),
857 if (S_ISDIR(entry.mode)) {
858 struct tree_desc sub;
859 struct pbase_tree_cache *tree;
860 const char *down = name+cmplen+1;
861 int downlen = name_cmp_len(down);
863 tree = pbase_tree_get(entry.sha1);
866 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
868 add_pbase_object(&sub, down, downlen, fullname);
869 pbase_tree_put(tree);
874 static unsigned *done_pbase_paths;
875 static int done_pbase_paths_num;
876 static int done_pbase_paths_alloc;
877 static int done_pbase_path_pos(unsigned hash)
880 int hi = done_pbase_paths_num;
882 int mi = (hi + lo) / 2;
883 if (done_pbase_paths[mi] == hash)
885 if (done_pbase_paths[mi] < hash)
893 static int check_pbase_path(unsigned hash)
895 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
899 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
900 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
901 done_pbase_paths = xrealloc(done_pbase_paths,
902 done_pbase_paths_alloc *
905 done_pbase_paths_num++;
906 if (pos < done_pbase_paths_num)
907 memmove(done_pbase_paths + pos + 1,
908 done_pbase_paths + pos,
909 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
910 done_pbase_paths[pos] = hash;
914 static void add_preferred_base_object(const char *name)
916 struct pbase_tree *it;
918 unsigned hash = name_hash(name);
920 if (!num_preferred_base || check_pbase_path(hash))
923 cmplen = name_cmp_len(name);
924 for (it = pbase_tree; it; it = it->next) {
926 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
929 struct tree_desc tree;
930 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
931 add_pbase_object(&tree, name, cmplen, name);
936 static void add_preferred_base(unsigned char *sha1)
938 struct pbase_tree *it;
941 unsigned char tree_sha1[20];
943 if (window <= num_preferred_base++)
946 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
950 for (it = pbase_tree; it; it = it->next) {
951 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
957 it = xcalloc(1, sizeof(*it));
958 it->next = pbase_tree;
961 hashcpy(it->pcache.sha1, tree_sha1);
962 it->pcache.tree_data = data;
963 it->pcache.tree_size = size;
966 static void cleanup_preferred_base(void)
968 struct pbase_tree *it;
974 struct pbase_tree *this = it;
976 free(this->pcache.tree_data);
980 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
981 if (!pbase_tree_cache[i])
983 free(pbase_tree_cache[i]->tree_data);
984 free(pbase_tree_cache[i]);
985 pbase_tree_cache[i] = NULL;
988 free(done_pbase_paths);
989 done_pbase_paths = NULL;
990 done_pbase_paths_num = done_pbase_paths_alloc = 0;
993 static void check_object(struct object_entry *entry)
995 if (entry->in_pack) {
996 struct packed_git *p = entry->in_pack;
997 struct pack_window *w_curs = NULL;
998 const unsigned char *base_ref = NULL;
999 struct object_entry *base_entry;
1000 unsigned long used, used_0;
1003 unsigned char *buf, c;
1005 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1008 * We want in_pack_type even if we do not reuse delta
1009 * since non-delta representations could still be reused.
1011 used = unpack_object_header_buffer(buf, avail,
1012 &entry->in_pack_type,
1018 * Determine if this is a delta and if so whether we can
1019 * reuse it or not. Otherwise let's find out as cheaply as
1020 * possible what the actual type and size for this object is.
1022 switch (entry->in_pack_type) {
1024 /* Not a delta hence we've already got all we need. */
1025 entry->type = entry->in_pack_type;
1026 entry->in_pack_header_size = used;
1027 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1029 unuse_pack(&w_curs);
1032 if (reuse_delta && !entry->preferred_base)
1033 base_ref = use_pack(p, &w_curs,
1034 entry->in_pack_offset + used, NULL);
1035 entry->in_pack_header_size = used + 20;
1038 buf = use_pack(p, &w_curs,
1039 entry->in_pack_offset + used, NULL);
1045 if (!ofs || MSB(ofs, 7)) {
1046 error("delta base offset overflow in pack for %s",
1047 sha1_to_hex(entry->idx.sha1));
1051 ofs = (ofs << 7) + (c & 127);
1053 ofs = entry->in_pack_offset - ofs;
1054 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1055 error("delta base offset out of bound for %s",
1056 sha1_to_hex(entry->idx.sha1));
1059 if (reuse_delta && !entry->preferred_base) {
1060 struct revindex_entry *revidx;
1061 revidx = find_pack_revindex(p, ofs);
1064 base_ref = nth_packed_object_sha1(p, revidx->nr);
1066 entry->in_pack_header_size = used + used_0;
1070 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1072 * If base_ref was set above that means we wish to
1073 * reuse delta data, and we even found that base
1074 * in the list of objects we want to pack. Goodie!
1076 * Depth value does not matter - find_deltas() will
1077 * never consider reused delta as the base object to
1078 * deltify other objects against, in order to avoid
1081 entry->type = entry->in_pack_type;
1082 entry->delta = base_entry;
1083 entry->delta_size = entry->size;
1084 entry->delta_sibling = base_entry->delta_child;
1085 base_entry->delta_child = entry;
1086 unuse_pack(&w_curs);
1092 * This must be a delta and we already know what the
1093 * final object type is. Let's extract the actual
1094 * object size from the delta header.
1096 entry->size = get_size_from_delta(p, &w_curs,
1097 entry->in_pack_offset + entry->in_pack_header_size);
1098 if (entry->size == 0)
1100 unuse_pack(&w_curs);
1105 * No choice but to fall back to the recursive delta walk
1106 * with sha1_object_info() to find about the object type
1110 unuse_pack(&w_curs);
1113 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1115 * The error condition is checked in prepare_pack(). This is
1116 * to permit a missing preferred base object to be ignored
1117 * as a preferred base. Doing so can result in a larger
1118 * pack file, but the transfer will still take place.
1122 static int pack_offset_sort(const void *_a, const void *_b)
1124 const struct object_entry *a = *(struct object_entry **)_a;
1125 const struct object_entry *b = *(struct object_entry **)_b;
1127 /* avoid filesystem trashing with loose objects */
1128 if (!a->in_pack && !b->in_pack)
1129 return hashcmp(a->idx.sha1, b->idx.sha1);
1131 if (a->in_pack < b->in_pack)
1133 if (a->in_pack > b->in_pack)
1135 return a->in_pack_offset < b->in_pack_offset ? -1 :
1136 (a->in_pack_offset > b->in_pack_offset);
1139 static void get_object_details(void)
1142 struct object_entry **sorted_by_offset;
1144 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1145 for (i = 0; i < nr_objects; i++)
1146 sorted_by_offset[i] = objects + i;
1147 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1149 for (i = 0; i < nr_objects; i++)
1150 check_object(sorted_by_offset[i]);
1152 free(sorted_by_offset);
1156 * We search for deltas in a list sorted by type, by filename hash, and then
1157 * by size, so that we see progressively smaller and smaller files.
1158 * That's because we prefer deltas to be from the bigger file
1159 * to the smaller -- deletes are potentially cheaper, but perhaps
1160 * more importantly, the bigger file is likely the more recent
1161 * one. The deepest deltas are therefore the oldest objects which are
1162 * less susceptible to be accessed often.
1164 static int type_size_sort(const void *_a, const void *_b)
1166 const struct object_entry *a = *(struct object_entry **)_a;
1167 const struct object_entry *b = *(struct object_entry **)_b;
1169 if (a->type > b->type)
1171 if (a->type < b->type)
1173 if (a->hash > b->hash)
1175 if (a->hash < b->hash)
1177 if (a->preferred_base > b->preferred_base)
1179 if (a->preferred_base < b->preferred_base)
1181 if (a->size > b->size)
1183 if (a->size < b->size)
1185 return a < b ? -1 : (a > b); /* newest first */
1189 struct object_entry *entry;
1191 struct delta_index *index;
1195 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1196 unsigned long delta_size)
1198 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1201 if (delta_size < cache_max_small_delta_size)
1204 /* cache delta, if objects are large enough compared to delta size */
1205 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1213 static pthread_mutex_t read_mutex;
1214 #define read_lock() pthread_mutex_lock(&read_mutex)
1215 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1217 static pthread_mutex_t cache_mutex;
1218 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1219 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1221 static pthread_mutex_t progress_mutex;
1222 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1223 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1227 #define read_lock() (void)0
1228 #define read_unlock() (void)0
1229 #define cache_lock() (void)0
1230 #define cache_unlock() (void)0
1231 #define progress_lock() (void)0
1232 #define progress_unlock() (void)0
1236 static int try_delta(struct unpacked *trg, struct unpacked *src,
1237 unsigned max_depth, unsigned long *mem_usage)
1239 struct object_entry *trg_entry = trg->entry;
1240 struct object_entry *src_entry = src->entry;
1241 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1243 enum object_type type;
1246 /* Don't bother doing diffs between different types */
1247 if (trg_entry->type != src_entry->type)
1251 * We do not bother to try a delta that we discarded
1252 * on an earlier try, but only when reusing delta data.
1254 if (reuse_delta && trg_entry->in_pack &&
1255 trg_entry->in_pack == src_entry->in_pack &&
1256 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1257 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1260 /* Let's not bust the allowed depth. */
1261 if (src->depth >= max_depth)
1264 /* Now some size filtering heuristics. */
1265 trg_size = trg_entry->size;
1266 if (!trg_entry->delta) {
1267 max_size = trg_size/2 - 20;
1270 max_size = trg_entry->delta_size;
1271 ref_depth = trg->depth;
1273 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1274 (max_depth - ref_depth + 1);
1277 src_size = src_entry->size;
1278 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1279 if (sizediff >= max_size)
1281 if (trg_size < src_size / 32)
1284 /* Load data if not already done */
1287 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1290 die("object %s cannot be read",
1291 sha1_to_hex(trg_entry->idx.sha1));
1293 die("object %s inconsistent object length (%lu vs %lu)",
1294 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1299 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1302 if (src_entry->preferred_base) {
1303 static int warned = 0;
1305 warning("object %s cannot be read",
1306 sha1_to_hex(src_entry->idx.sha1));
1308 * Those objects are not included in the
1309 * resulting pack. Be resilient and ignore
1310 * them if they can't be read, in case the
1311 * pack could be created nevertheless.
1315 die("object %s cannot be read",
1316 sha1_to_hex(src_entry->idx.sha1));
1319 die("object %s inconsistent object length (%lu vs %lu)",
1320 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1324 src->index = create_delta_index(src->data, src_size);
1326 static int warned = 0;
1328 warning("suboptimal pack - out of memory");
1331 *mem_usage += sizeof_delta_index(src->index);
1334 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1338 if (trg_entry->delta) {
1339 /* Prefer only shallower same-sized deltas. */
1340 if (delta_size == trg_entry->delta_size &&
1341 src->depth + 1 >= trg->depth) {
1348 * Handle memory allocation outside of the cache
1349 * accounting lock. Compiler will optimize the strangeness
1350 * away when NO_PTHREADS is defined.
1352 free(trg_entry->delta_data);
1354 if (trg_entry->delta_data) {
1355 delta_cache_size -= trg_entry->delta_size;
1356 trg_entry->delta_data = NULL;
1358 if (delta_cacheable(src_size, trg_size, delta_size)) {
1359 delta_cache_size += delta_size;
1361 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1367 trg_entry->delta = src_entry;
1368 trg_entry->delta_size = delta_size;
1369 trg->depth = src->depth + 1;
1374 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1376 struct object_entry *child = me->delta_child;
1379 unsigned int c = check_delta_limit(child, n + 1);
1382 child = child->delta_sibling;
1387 static unsigned long free_unpacked(struct unpacked *n)
1389 unsigned long freed_mem = sizeof_delta_index(n->index);
1390 free_delta_index(n->index);
1393 freed_mem += n->entry->size;
1402 static void find_deltas(struct object_entry **list, unsigned *list_size,
1403 int window, int depth, unsigned *processed)
1405 uint32_t i, idx = 0, count = 0;
1406 struct unpacked *array;
1407 unsigned long mem_usage = 0;
1409 array = xcalloc(window, sizeof(struct unpacked));
1412 struct object_entry *entry;
1413 struct unpacked *n = array + idx;
1414 int j, max_depth, best_base = -1;
1423 if (!entry->preferred_base) {
1425 display_progress(progress_state, *processed);
1429 mem_usage -= free_unpacked(n);
1432 while (window_memory_limit &&
1433 mem_usage > window_memory_limit &&
1435 uint32_t tail = (idx + window - count) % window;
1436 mem_usage -= free_unpacked(array + tail);
1440 /* We do not compute delta to *create* objects we are not
1443 if (entry->preferred_base)
1447 * If the current object is at pack edge, take the depth the
1448 * objects that depend on the current object into account
1449 * otherwise they would become too deep.
1452 if (entry->delta_child) {
1453 max_depth -= check_delta_limit(entry, 0);
1461 uint32_t other_idx = idx + j;
1463 if (other_idx >= window)
1464 other_idx -= window;
1465 m = array + other_idx;
1468 ret = try_delta(n, m, max_depth, &mem_usage);
1472 best_base = other_idx;
1476 * If we decided to cache the delta data, then it is best
1477 * to compress it right away. First because we have to do
1478 * it anyway, and doing it here while we're threaded will
1479 * save a lot of time in the non threaded write phase,
1480 * as well as allow for caching more deltas within
1481 * the same cache size limit.
1483 * But only if not writing to stdout, since in that case
1484 * the network is most likely throttling writes anyway,
1485 * and therefore it is best to go to the write phase ASAP
1486 * instead, as we can afford spending more time compressing
1487 * between writes at that moment.
1489 if (entry->delta_data && !pack_to_stdout) {
1490 entry->z_delta_size = do_compress(&entry->delta_data,
1493 delta_cache_size -= entry->delta_size;
1494 delta_cache_size += entry->z_delta_size;
1498 /* if we made n a delta, and if n is already at max
1499 * depth, leaving it in the window is pointless. we
1500 * should evict it first.
1502 if (entry->delta && max_depth <= n->depth)
1506 * Move the best delta base up in the window, after the
1507 * currently deltified object, to keep it longer. It will
1508 * be the first base object to be attempted next.
1511 struct unpacked swap = array[best_base];
1512 int dist = (window + idx - best_base) % window;
1513 int dst = best_base;
1515 int src = (dst + 1) % window;
1516 array[dst] = array[src];
1524 if (count + 1 < window)
1530 for (i = 0; i < window; ++i) {
1531 free_delta_index(array[i].index);
1532 free(array[i].data);
1539 static void try_to_free_from_threads(size_t size)
1542 release_pack_memory(size, -1);
1546 try_to_free_t old_try_to_free_routine;
1549 * The main thread waits on the condition that (at least) one of the workers
1550 * has stopped working (which is indicated in the .working member of
1551 * struct thread_params).
1552 * When a work thread has completed its work, it sets .working to 0 and
1553 * signals the main thread and waits on the condition that .data_ready
1557 struct thread_params {
1559 struct object_entry **list;
1566 pthread_mutex_t mutex;
1567 pthread_cond_t cond;
1568 unsigned *processed;
1571 static pthread_cond_t progress_cond;
1574 * Mutex and conditional variable can't be statically-initialized on Windows.
1576 static void init_threaded_search(void)
1578 init_recursive_mutex(&read_mutex);
1579 pthread_mutex_init(&cache_mutex, NULL);
1580 pthread_mutex_init(&progress_mutex, NULL);
1581 pthread_cond_init(&progress_cond, NULL);
1582 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1585 static void cleanup_threaded_search(void)
1587 set_try_to_free_routine(old_try_to_free_routine);
1588 pthread_cond_destroy(&progress_cond);
1589 pthread_mutex_destroy(&read_mutex);
1590 pthread_mutex_destroy(&cache_mutex);
1591 pthread_mutex_destroy(&progress_mutex);
1594 static void *threaded_find_deltas(void *arg)
1596 struct thread_params *me = arg;
1598 while (me->remaining) {
1599 find_deltas(me->list, &me->remaining,
1600 me->window, me->depth, me->processed);
1604 pthread_cond_signal(&progress_cond);
1608 * We must not set ->data_ready before we wait on the
1609 * condition because the main thread may have set it to 1
1610 * before we get here. In order to be sure that new
1611 * work is available if we see 1 in ->data_ready, it
1612 * was initialized to 0 before this thread was spawned
1613 * and we reset it to 0 right away.
1615 pthread_mutex_lock(&me->mutex);
1616 while (!me->data_ready)
1617 pthread_cond_wait(&me->cond, &me->mutex);
1619 pthread_mutex_unlock(&me->mutex);
1621 /* leave ->working 1 so that this doesn't get more work assigned */
1625 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1626 int window, int depth, unsigned *processed)
1628 struct thread_params *p;
1629 int i, ret, active_threads = 0;
1631 init_threaded_search();
1633 if (!delta_search_threads) /* --threads=0 means autodetect */
1634 delta_search_threads = online_cpus();
1635 if (delta_search_threads <= 1) {
1636 find_deltas(list, &list_size, window, depth, processed);
1637 cleanup_threaded_search();
1640 if (progress > pack_to_stdout)
1641 fprintf(stderr, "Delta compression using up to %d threads.\n",
1642 delta_search_threads);
1643 p = xcalloc(delta_search_threads, sizeof(*p));
1645 /* Partition the work amongst work threads. */
1646 for (i = 0; i < delta_search_threads; i++) {
1647 unsigned sub_size = list_size / (delta_search_threads - i);
1649 /* don't use too small segments or no deltas will be found */
1650 if (sub_size < 2*window && i+1 < delta_search_threads)
1653 p[i].window = window;
1655 p[i].processed = processed;
1657 p[i].data_ready = 0;
1659 /* try to split chunks on "path" boundaries */
1660 while (sub_size && sub_size < list_size &&
1661 list[sub_size]->hash &&
1662 list[sub_size]->hash == list[sub_size-1]->hash)
1666 p[i].list_size = sub_size;
1667 p[i].remaining = sub_size;
1670 list_size -= sub_size;
1673 /* Start work threads. */
1674 for (i = 0; i < delta_search_threads; i++) {
1675 if (!p[i].list_size)
1677 pthread_mutex_init(&p[i].mutex, NULL);
1678 pthread_cond_init(&p[i].cond, NULL);
1679 ret = pthread_create(&p[i].thread, NULL,
1680 threaded_find_deltas, &p[i]);
1682 die("unable to create thread: %s", strerror(ret));
1687 * Now let's wait for work completion. Each time a thread is done
1688 * with its work, we steal half of the remaining work from the
1689 * thread with the largest number of unprocessed objects and give
1690 * it to that newly idle thread. This ensure good load balancing
1691 * until the remaining object list segments are simply too short
1692 * to be worth splitting anymore.
1694 while (active_threads) {
1695 struct thread_params *target = NULL;
1696 struct thread_params *victim = NULL;
1697 unsigned sub_size = 0;
1701 for (i = 0; !target && i < delta_search_threads; i++)
1706 pthread_cond_wait(&progress_cond, &progress_mutex);
1709 for (i = 0; i < delta_search_threads; i++)
1710 if (p[i].remaining > 2*window &&
1711 (!victim || victim->remaining < p[i].remaining))
1714 sub_size = victim->remaining / 2;
1715 list = victim->list + victim->list_size - sub_size;
1716 while (sub_size && list[0]->hash &&
1717 list[0]->hash == list[-1]->hash) {
1723 * It is possible for some "paths" to have
1724 * so many objects that no hash boundary
1725 * might be found. Let's just steal the
1726 * exact half in that case.
1728 sub_size = victim->remaining / 2;
1731 target->list = list;
1732 victim->list_size -= sub_size;
1733 victim->remaining -= sub_size;
1735 target->list_size = sub_size;
1736 target->remaining = sub_size;
1737 target->working = 1;
1740 pthread_mutex_lock(&target->mutex);
1741 target->data_ready = 1;
1742 pthread_cond_signal(&target->cond);
1743 pthread_mutex_unlock(&target->mutex);
1746 pthread_join(target->thread, NULL);
1747 pthread_cond_destroy(&target->cond);
1748 pthread_mutex_destroy(&target->mutex);
1752 cleanup_threaded_search();
1757 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1760 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1762 unsigned char peeled[20];
1764 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1765 !peel_ref(path, peeled) && /* peelable? */
1766 !is_null_sha1(peeled) && /* annotated tag? */
1767 locate_object_entry(peeled)) /* object packed? */
1768 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1772 static void prepare_pack(int window, int depth)
1774 struct object_entry **delta_list;
1775 uint32_t i, nr_deltas;
1778 get_object_details();
1781 * If we're locally repacking then we need to be doubly careful
1782 * from now on in order to make sure no stealth corruption gets
1783 * propagated to the new pack. Clients receiving streamed packs
1784 * should validate everything they get anyway so no need to incur
1785 * the additional cost here in that case.
1787 if (!pack_to_stdout)
1788 do_check_packed_object_crc = 1;
1790 if (!nr_objects || !window || !depth)
1793 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1796 for (i = 0; i < nr_objects; i++) {
1797 struct object_entry *entry = objects + i;
1800 /* This happens if we decided to reuse existing
1801 * delta from a pack. "reuse_delta &&" is implied.
1805 if (entry->size < 50)
1808 if (entry->no_try_delta)
1811 if (!entry->preferred_base) {
1813 if (entry->type < 0)
1814 die("unable to get type of object %s",
1815 sha1_to_hex(entry->idx.sha1));
1817 if (entry->type < 0) {
1819 * This object is not found, but we
1820 * don't have to include it anyway.
1826 delta_list[n++] = entry;
1829 if (nr_deltas && n > 1) {
1830 unsigned nr_done = 0;
1832 progress_state = start_progress("Compressing objects",
1834 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1835 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1836 stop_progress(&progress_state);
1837 if (nr_done != nr_deltas)
1838 die("inconsistency with delta count");
1843 static int git_pack_config(const char *k, const char *v, void *cb)
1845 if (!strcmp(k, "pack.window")) {
1846 window = git_config_int(k, v);
1849 if (!strcmp(k, "pack.windowmemory")) {
1850 window_memory_limit = git_config_ulong(k, v);
1853 if (!strcmp(k, "pack.depth")) {
1854 depth = git_config_int(k, v);
1857 if (!strcmp(k, "pack.compression")) {
1858 int level = git_config_int(k, v);
1860 level = Z_DEFAULT_COMPRESSION;
1861 else if (level < 0 || level > Z_BEST_COMPRESSION)
1862 die("bad pack compression level %d", level);
1863 pack_compression_level = level;
1864 pack_compression_seen = 1;
1867 if (!strcmp(k, "pack.deltacachesize")) {
1868 max_delta_cache_size = git_config_int(k, v);
1871 if (!strcmp(k, "pack.deltacachelimit")) {
1872 cache_max_small_delta_size = git_config_int(k, v);
1875 if (!strcmp(k, "pack.threads")) {
1876 delta_search_threads = git_config_int(k, v);
1877 if (delta_search_threads < 0)
1878 die("invalid number of threads specified (%d)",
1879 delta_search_threads);
1881 if (delta_search_threads != 1)
1882 warning("no threads support, ignoring %s", k);
1886 if (!strcmp(k, "pack.indexversion")) {
1887 pack_idx_default_version = git_config_int(k, v);
1888 if (pack_idx_default_version > 2)
1889 die("bad pack.indexversion=%"PRIu32,
1890 pack_idx_default_version);
1893 if (!strcmp(k, "pack.packsizelimit")) {
1894 pack_size_limit_cfg = git_config_ulong(k, v);
1897 return git_default_config(k, v, cb);
1900 static void read_object_list_from_stdin(void)
1902 char line[40 + 1 + PATH_MAX + 2];
1903 unsigned char sha1[20];
1906 if (!fgets(line, sizeof(line), stdin)) {
1910 die("fgets returned NULL, not EOF, not error!");
1916 if (line[0] == '-') {
1917 if (get_sha1_hex(line+1, sha1))
1918 die("expected edge sha1, got garbage:\n %s",
1920 add_preferred_base(sha1);
1923 if (get_sha1_hex(line, sha1))
1924 die("expected sha1, got garbage:\n %s", line);
1926 add_preferred_base_object(line+41);
1927 add_object_entry(sha1, 0, line+41, 0);
1931 #define OBJECT_ADDED (1u<<20)
1933 static void show_commit(struct commit *commit, void *data)
1935 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1936 commit->object.flags |= OBJECT_ADDED;
1939 static void show_object(struct object *obj, const struct name_path *path, const char *last)
1941 char *name = path_name(path, last);
1943 add_preferred_base_object(name);
1944 add_object_entry(obj->sha1, obj->type, name, 0);
1945 obj->flags |= OBJECT_ADDED;
1948 * We will have generated the hash from the name,
1949 * but not saved a pointer to it - we can free it
1954 static void show_edge(struct commit *commit)
1956 add_preferred_base(commit->object.sha1);
1959 struct in_pack_object {
1961 struct object *object;
1967 struct in_pack_object *array;
1970 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1972 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1973 in_pack->array[in_pack->nr].object = object;
1978 * Compare the objects in the offset order, in order to emulate the
1979 * "git rev-list --objects" output that produced the pack originally.
1981 static int ofscmp(const void *a_, const void *b_)
1983 struct in_pack_object *a = (struct in_pack_object *)a_;
1984 struct in_pack_object *b = (struct in_pack_object *)b_;
1986 if (a->offset < b->offset)
1988 else if (a->offset > b->offset)
1991 return hashcmp(a->object->sha1, b->object->sha1);
1994 static void add_objects_in_unpacked_packs(struct rev_info *revs)
1996 struct packed_git *p;
1997 struct in_pack in_pack;
2000 memset(&in_pack, 0, sizeof(in_pack));
2002 for (p = packed_git; p; p = p->next) {
2003 const unsigned char *sha1;
2006 if (!p->pack_local || p->pack_keep)
2008 if (open_pack_index(p))
2009 die("cannot open pack index");
2011 ALLOC_GROW(in_pack.array,
2012 in_pack.nr + p->num_objects,
2015 for (i = 0; i < p->num_objects; i++) {
2016 sha1 = nth_packed_object_sha1(p, i);
2017 o = lookup_unknown_object(sha1);
2018 if (!(o->flags & OBJECT_ADDED))
2019 mark_in_pack_object(o, p, &in_pack);
2020 o->flags |= OBJECT_ADDED;
2025 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2027 for (i = 0; i < in_pack.nr; i++) {
2028 struct object *o = in_pack.array[i].object;
2029 add_object_entry(o->sha1, o->type, "", 0);
2032 free(in_pack.array);
2035 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2037 static struct packed_git *last_found = (void *)1;
2038 struct packed_git *p;
2040 p = (last_found != (void *)1) ? last_found : packed_git;
2043 if ((!p->pack_local || p->pack_keep) &&
2044 find_pack_entry_one(sha1, p)) {
2048 if (p == last_found)
2052 if (p == last_found)
2058 static void loosen_unused_packed_objects(struct rev_info *revs)
2060 struct packed_git *p;
2062 const unsigned char *sha1;
2064 for (p = packed_git; p; p = p->next) {
2065 if (!p->pack_local || p->pack_keep)
2068 if (open_pack_index(p))
2069 die("cannot open pack index");
2071 for (i = 0; i < p->num_objects; i++) {
2072 sha1 = nth_packed_object_sha1(p, i);
2073 if (!locate_object_entry(sha1) &&
2074 !has_sha1_pack_kept_or_nonlocal(sha1))
2075 if (force_object_loose(sha1, p->mtime))
2076 die("unable to force loose object");
2081 static void get_object_list(int ac, const char **av)
2083 struct rev_info revs;
2087 init_revisions(&revs, NULL);
2088 save_commit_buffer = 0;
2089 setup_revisions(ac, av, &revs, NULL);
2091 while (fgets(line, sizeof(line), stdin) != NULL) {
2092 int len = strlen(line);
2093 if (len && line[len - 1] == '\n')
2098 if (!strcmp(line, "--not")) {
2099 flags ^= UNINTERESTING;
2102 die("not a rev '%s'", line);
2104 if (handle_revision_arg(line, &revs, flags, 1))
2105 die("bad revision '%s'", line);
2108 if (prepare_revision_walk(&revs))
2109 die("revision walk setup failed");
2110 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2111 traverse_commit_list(&revs, show_commit, show_object, NULL);
2113 if (keep_unreachable)
2114 add_objects_in_unpacked_packs(&revs);
2115 if (unpack_unreachable)
2116 loosen_unused_packed_objects(&revs);
2119 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2121 int use_internal_rev_list = 0;
2123 int all_progress_implied = 0;
2126 int rp_ac_alloc = 64;
2129 read_replace_refs = 0;
2131 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2133 rp_av[0] = "pack-objects";
2134 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2137 git_config(git_pack_config, NULL);
2138 if (!pack_compression_seen && core_compression_seen)
2139 pack_compression_level = core_compression_level;
2141 progress = isatty(2);
2142 for (i = 1; i < argc; i++) {
2143 const char *arg = argv[i];
2148 if (!strcmp("--non-empty", arg)) {
2152 if (!strcmp("--local", arg)) {
2156 if (!strcmp("--incremental", arg)) {
2160 if (!strcmp("--honor-pack-keep", arg)) {
2161 ignore_packed_keep = 1;
2164 if (!prefixcmp(arg, "--compression=")) {
2166 int level = strtoul(arg+14, &end, 0);
2167 if (!arg[14] || *end)
2170 level = Z_DEFAULT_COMPRESSION;
2171 else if (level < 0 || level > Z_BEST_COMPRESSION)
2172 die("bad pack compression level %d", level);
2173 pack_compression_level = level;
2176 if (!prefixcmp(arg, "--max-pack-size=")) {
2177 pack_size_limit_cfg = 0;
2178 if (!git_parse_ulong(arg+16, &pack_size_limit))
2182 if (!prefixcmp(arg, "--window=")) {
2184 window = strtoul(arg+9, &end, 0);
2185 if (!arg[9] || *end)
2189 if (!prefixcmp(arg, "--window-memory=")) {
2190 if (!git_parse_ulong(arg+16, &window_memory_limit))
2194 if (!prefixcmp(arg, "--threads=")) {
2196 delta_search_threads = strtoul(arg+10, &end, 0);
2197 if (!arg[10] || *end || delta_search_threads < 0)
2200 if (delta_search_threads != 1)
2201 warning("no threads support, "
2202 "ignoring %s", arg);
2206 if (!prefixcmp(arg, "--depth=")) {
2208 depth = strtoul(arg+8, &end, 0);
2209 if (!arg[8] || *end)
2213 if (!strcmp("--progress", arg)) {
2217 if (!strcmp("--all-progress", arg)) {
2221 if (!strcmp("--all-progress-implied", arg)) {
2222 all_progress_implied = 1;
2225 if (!strcmp("-q", arg)) {
2229 if (!strcmp("--no-reuse-delta", arg)) {
2233 if (!strcmp("--no-reuse-object", arg)) {
2234 reuse_object = reuse_delta = 0;
2237 if (!strcmp("--delta-base-offset", arg)) {
2238 allow_ofs_delta = 1;
2241 if (!strcmp("--stdout", arg)) {
2245 if (!strcmp("--revs", arg)) {
2246 use_internal_rev_list = 1;
2249 if (!strcmp("--keep-unreachable", arg)) {
2250 keep_unreachable = 1;
2253 if (!strcmp("--unpack-unreachable", arg)) {
2254 unpack_unreachable = 1;
2257 if (!strcmp("--include-tag", arg)) {
2261 if (!strcmp("--unpacked", arg) ||
2262 !strcmp("--reflog", arg) ||
2263 !strcmp("--all", arg)) {
2264 use_internal_rev_list = 1;
2265 if (rp_ac >= rp_ac_alloc - 1) {
2266 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2267 rp_av = xrealloc(rp_av,
2268 rp_ac_alloc * sizeof(*rp_av));
2270 rp_av[rp_ac++] = arg;
2273 if (!strcmp("--thin", arg)) {
2274 use_internal_rev_list = 1;
2276 rp_av[1] = "--objects-edge";
2279 if (!prefixcmp(arg, "--index-version=")) {
2281 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2282 if (pack_idx_default_version > 2)
2285 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2286 if (*c || pack_idx_off32_limit & 0x80000000)
2290 if (!strcmp(arg, "--keep-true-parents")) {
2291 grafts_replace_parents = 0;
2297 /* Traditionally "pack-objects [options] base extra" failed;
2298 * we would however want to take refs parameter that would
2299 * have been given to upstream rev-list ourselves, which means
2300 * we somehow want to say what the base name is. So the
2303 * pack-objects [options] base <refs...>
2305 * in other words, we would treat the first non-option as the
2306 * base_name and send everything else to the internal revision
2310 if (!pack_to_stdout)
2311 base_name = argv[i++];
2313 if (pack_to_stdout != !base_name)
2316 if (!pack_to_stdout && !pack_size_limit)
2317 pack_size_limit = pack_size_limit_cfg;
2318 if (pack_to_stdout && pack_size_limit)
2319 die("--max-pack-size cannot be used to build a pack for transfer.");
2320 if (pack_size_limit && pack_size_limit < 1024*1024) {
2321 warning("minimum pack size limit is 1 MiB");
2322 pack_size_limit = 1024*1024;
2325 if (!pack_to_stdout && thin)
2326 die("--thin cannot be used to build an indexable pack.");
2328 if (keep_unreachable && unpack_unreachable)
2329 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2331 if (progress && all_progress_implied)
2334 prepare_packed_git();
2337 progress_state = start_progress("Counting objects", 0);
2338 if (!use_internal_rev_list)
2339 read_object_list_from_stdin();
2341 rp_av[rp_ac] = NULL;
2342 get_object_list(rp_ac, rp_av);
2344 cleanup_preferred_base();
2345 if (include_tag && nr_result)
2346 for_each_ref(add_ref_tag, NULL);
2347 stop_progress(&progress_state);
2349 if (non_empty && !nr_result)
2352 prepare_pack(window, depth);
2355 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2356 " reused %"PRIu32" (delta %"PRIu32")\n",
2357 written, written_delta, reused, reused_delta);