11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
16 #include "list-objects.h"
20 #ifdef THREADED_DELTA_SEARCH
21 #include "thread-utils.h"
25 static const char pack_usage[] = "\
26 git-pack-objects [{ -q | --progress | --all-progress }] \n\
27 [--max-pack-size=N] [--local] [--incremental] \n\
28 [--window=N] [--window-memory=N] [--depth=N] \n\
29 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
30 [--threads=N] [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
31 [--stdout | base-name] [--include-tag] \n\
32 [--keep-unreachable | --unpack-unreachable] \n\
33 [<ref-list | <object-list]";
36 struct pack_idx_entry idx;
37 unsigned long size; /* uncompressed size */
38 struct packed_git *in_pack; /* already in pack */
40 struct object_entry *delta; /* delta base object */
41 struct object_entry *delta_child; /* deltified objects who bases me */
42 struct object_entry *delta_sibling; /* other deltified objects who
43 * uses the same base as me
45 void *delta_data; /* cached delta (uncompressed) */
46 unsigned long delta_size; /* delta data size (uncompressed) */
47 unsigned long z_delta_size; /* delta data size (compressed) */
48 unsigned int hash; /* name hint hash */
49 enum object_type type;
50 enum object_type in_pack_type; /* could be delta */
51 unsigned char in_pack_header_size;
52 unsigned char preferred_base; /* we do not pack this, but is available
53 * to be used as the base object to delta
56 unsigned char no_try_delta;
60 * Objects we are going to pack are collected in objects array (dynamically
61 * expanded). nr_objects & nr_alloc controls this array. They are stored
62 * in the order we see -- typically rev-list --objects order that gives us
63 * nice "minimum seek" order.
65 static struct object_entry *objects;
66 static struct pack_idx_entry **written_list;
67 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
70 static int reuse_delta = 1, reuse_object = 1;
71 static int keep_unreachable, unpack_unreachable, include_tag;
73 static int incremental;
74 static int allow_ofs_delta;
75 static const char *base_name;
76 static int progress = 1;
77 static int window = 10;
78 static uint32_t pack_size_limit, pack_size_limit_cfg;
79 static int depth = 50;
80 static int delta_search_threads = 1;
81 static int pack_to_stdout;
82 static int num_preferred_base;
83 static struct progress *progress_state;
84 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
85 static int pack_compression_seen;
87 static unsigned long delta_cache_size = 0;
88 static unsigned long max_delta_cache_size = 0;
89 static unsigned long cache_max_small_delta_size = 1000;
91 static unsigned long window_memory_limit = 0;
94 * The object names in objects array are hashed with this hashtable,
95 * to help looking up the entry by object name.
96 * This hashtable is built after all the objects are seen.
98 static int *object_ix;
99 static int object_ix_hashsz;
104 static uint32_t written, written_delta;
105 static uint32_t reused, reused_delta;
108 static void *get_delta(struct object_entry *entry)
110 unsigned long size, base_size, delta_size;
111 void *buf, *base_buf, *delta_buf;
112 enum object_type type;
114 buf = read_sha1_file(entry->idx.sha1, &type, &size);
116 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
117 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
119 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
120 delta_buf = diff_delta(base_buf, base_size,
121 buf, size, &delta_size, 0);
122 if (!delta_buf || delta_size != entry->delta_size)
123 die("delta size changed");
129 static unsigned long do_compress(void **pptr, unsigned long size)
133 unsigned long maxsize;
135 memset(&stream, 0, sizeof(stream));
136 deflateInit(&stream, pack_compression_level);
137 maxsize = deflateBound(&stream, size);
140 out = xmalloc(maxsize);
144 stream.avail_in = size;
145 stream.next_out = out;
146 stream.avail_out = maxsize;
147 while (deflate(&stream, Z_FINISH) == Z_OK)
152 return stream.total_out;
156 * The per-object header is a pretty dense thing, which is
157 * - first byte: low four bits are "size", then three bits of "type",
158 * and the high bit is "size continues".
159 * - each byte afterwards: low seven bits are size continuation,
160 * with the high bit being "size continues"
162 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
167 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
168 die("bad type %d", type);
170 c = (type << 4) | (size & 15);
183 * we are going to reuse the existing object data as is. make
184 * sure it is not corrupt.
186 static int check_pack_inflate(struct packed_git *p,
187 struct pack_window **w_curs,
190 unsigned long expect)
193 unsigned char fakebuf[4096], *in;
196 memset(&stream, 0, sizeof(stream));
197 inflateInit(&stream);
199 in = use_pack(p, w_curs, offset, &stream.avail_in);
201 stream.next_out = fakebuf;
202 stream.avail_out = sizeof(fakebuf);
203 st = inflate(&stream, Z_FINISH);
204 offset += stream.next_in - in;
205 } while (st == Z_OK || st == Z_BUF_ERROR);
207 return (st == Z_STREAM_END &&
208 stream.total_out == expect &&
209 stream.total_in == len) ? 0 : -1;
212 static void copy_pack_data(struct sha1file *f,
213 struct packed_git *p,
214 struct pack_window **w_curs,
222 in = use_pack(p, w_curs, offset, &avail);
224 avail = (unsigned int)len;
225 sha1write(f, in, avail);
231 static unsigned long write_object(struct sha1file *f,
232 struct object_entry *entry,
235 unsigned long size, limit, datalen;
237 unsigned char header[10], dheader[10];
239 enum object_type type;
240 int usable_delta, to_reuse;
247 /* write limit if limited packsize and not first object */
248 limit = pack_size_limit && nr_written ?
249 pack_size_limit - write_offset : 0;
252 usable_delta = 0; /* no delta */
253 else if (!pack_size_limit)
254 usable_delta = 1; /* unlimited packfile */
255 else if (entry->delta->idx.offset == (off_t)-1)
256 usable_delta = 0; /* base was written to another pack */
257 else if (entry->delta->idx.offset)
258 usable_delta = 1; /* base already exists in this pack */
260 usable_delta = 0; /* base could end up in another pack */
263 to_reuse = 0; /* explicit */
264 else if (!entry->in_pack)
265 to_reuse = 0; /* can't reuse what we don't have */
266 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
267 /* check_object() decided it for us ... */
268 to_reuse = usable_delta;
269 /* ... but pack split may override that */
270 else if (type != entry->in_pack_type)
271 to_reuse = 0; /* pack has delta which is unusable */
272 else if (entry->delta)
273 to_reuse = 0; /* we want to pack afresh */
275 to_reuse = 1; /* we have it in-pack undeltified,
276 * and we do not need to deltify it.
281 buf = read_sha1_file(entry->idx.sha1, &type, &size);
283 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
285 * make sure no cached delta data remains from a
286 * previous attempt before a pack split occured.
288 free(entry->delta_data);
289 entry->delta_data = NULL;
290 entry->z_delta_size = 0;
291 } else if (entry->delta_data) {
292 size = entry->delta_size;
293 buf = entry->delta_data;
294 entry->delta_data = NULL;
295 type = (allow_ofs_delta && entry->delta->idx.offset) ?
296 OBJ_OFS_DELTA : OBJ_REF_DELTA;
298 buf = get_delta(entry);
299 size = entry->delta_size;
300 type = (allow_ofs_delta && entry->delta->idx.offset) ?
301 OBJ_OFS_DELTA : OBJ_REF_DELTA;
304 if (entry->z_delta_size)
305 datalen = entry->z_delta_size;
307 datalen = do_compress(&buf, size);
310 * The object header is a byte of 'type' followed by zero or
311 * more bytes of length.
313 hdrlen = encode_header(type, size, header);
315 if (type == OBJ_OFS_DELTA) {
317 * Deltas with relative base contain an additional
318 * encoding of the relative offset for the delta
319 * base from this object's position in the pack.
321 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
322 unsigned pos = sizeof(dheader) - 1;
323 dheader[pos] = ofs & 127;
325 dheader[--pos] = 128 | (--ofs & 127);
326 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
330 sha1write(f, header, hdrlen);
331 sha1write(f, dheader + pos, sizeof(dheader) - pos);
332 hdrlen += sizeof(dheader) - pos;
333 } else if (type == OBJ_REF_DELTA) {
335 * Deltas with a base reference contain
336 * an additional 20 bytes for the base sha1.
338 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
342 sha1write(f, header, hdrlen);
343 sha1write(f, entry->delta->idx.sha1, 20);
346 if (limit && hdrlen + datalen + 20 >= limit) {
350 sha1write(f, header, hdrlen);
352 sha1write(f, buf, datalen);
356 struct packed_git *p = entry->in_pack;
357 struct pack_window *w_curs = NULL;
358 struct revindex_entry *revidx;
362 type = (allow_ofs_delta && entry->delta->idx.offset) ?
363 OBJ_OFS_DELTA : OBJ_REF_DELTA;
366 hdrlen = encode_header(type, entry->size, header);
367 offset = entry->in_pack_offset;
368 revidx = find_pack_revindex(p, offset);
369 datalen = revidx[1].offset - offset;
370 if (!pack_to_stdout && p->index_version > 1 &&
371 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
372 die("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
373 offset += entry->in_pack_header_size;
374 datalen -= entry->in_pack_header_size;
375 if (type == OBJ_OFS_DELTA) {
376 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
377 unsigned pos = sizeof(dheader) - 1;
378 dheader[pos] = ofs & 127;
380 dheader[--pos] = 128 | (--ofs & 127);
381 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
383 sha1write(f, header, hdrlen);
384 sha1write(f, dheader + pos, sizeof(dheader) - pos);
385 hdrlen += sizeof(dheader) - pos;
386 } else if (type == OBJ_REF_DELTA) {
387 if (limit && hdrlen + 20 + datalen + 20 >= limit)
389 sha1write(f, header, hdrlen);
390 sha1write(f, entry->delta->idx.sha1, 20);
393 if (limit && hdrlen + datalen + 20 >= limit)
395 sha1write(f, header, hdrlen);
398 if (!pack_to_stdout && p->index_version == 1 &&
399 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
400 die("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
401 copy_pack_data(f, p, &w_curs, offset, datalen);
409 entry->idx.crc32 = crc32_end(f);
410 return hdrlen + datalen;
413 static off_t write_one(struct sha1file *f,
414 struct object_entry *e,
419 /* offset is non zero if object is written already. */
420 if (e->idx.offset || e->preferred_base)
423 /* if we are deltified, write out base object first. */
425 offset = write_one(f, e->delta, offset);
430 e->idx.offset = offset;
431 size = write_object(f, e, offset);
436 written_list[nr_written++] = &e->idx;
438 /* make sure off_t is sufficiently large not to wrap */
439 if (offset > offset + size)
440 die("pack too large for current definition of off_t");
441 return offset + size;
444 /* forward declaration for write_pack_file */
445 static int adjust_perm(const char *path, mode_t mode);
447 static void write_pack_file(void)
451 off_t offset, offset_one, last_obj_offset = 0;
452 struct pack_header hdr;
453 uint32_t nr_remaining = nr_result;
454 time_t last_mtime = 0;
456 if (progress > pack_to_stdout)
457 progress_state = start_progress("Writing objects", nr_result);
458 written_list = xmalloc(nr_objects * sizeof(*written_list));
461 unsigned char sha1[20];
462 char *pack_tmp_name = NULL;
464 if (pack_to_stdout) {
465 f = sha1fd_throughput(1, "<stdout>", progress_state);
467 char tmpname[PATH_MAX];
469 snprintf(tmpname, sizeof(tmpname),
470 "%s/tmp_pack_XXXXXX", get_object_directory());
471 fd = xmkstemp(tmpname);
472 pack_tmp_name = xstrdup(tmpname);
473 f = sha1fd(fd, pack_tmp_name);
476 hdr.hdr_signature = htonl(PACK_SIGNATURE);
477 hdr.hdr_version = htonl(PACK_VERSION);
478 hdr.hdr_entries = htonl(nr_remaining);
479 sha1write(f, &hdr, sizeof(hdr));
480 offset = sizeof(hdr);
482 for (; i < nr_objects; i++) {
483 last_obj_offset = offset;
484 offset_one = write_one(f, objects + i, offset);
488 display_progress(progress_state, written);
492 * Did we write the wrong # entries in the header?
493 * If so, rewrite it like in fast-import
495 if (pack_to_stdout) {
496 sha1close(f, sha1, CSUM_CLOSE);
497 } else if (nr_written == nr_remaining) {
498 sha1close(f, sha1, CSUM_FSYNC);
500 int fd = sha1close(f, NULL, 0);
501 fixup_pack_header_footer(fd, sha1, pack_tmp_name, nr_written);
502 fsync_or_die(fd, pack_tmp_name);
506 if (!pack_to_stdout) {
507 mode_t mode = umask(0);
509 char *idx_tmp_name, tmpname[PATH_MAX];
514 idx_tmp_name = write_idx_file(NULL, written_list,
517 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
518 base_name, sha1_to_hex(sha1));
519 if (adjust_perm(pack_tmp_name, mode))
520 die("unable to make temporary pack file readable: %s",
522 if (rename(pack_tmp_name, tmpname))
523 die("unable to rename temporary pack file: %s",
527 * Packs are runtime accessed in their mtime
528 * order since newer packs are more likely to contain
529 * younger objects. So if we are creating multiple
530 * packs then we should modify the mtime of later ones
531 * to preserve this property.
533 if (stat(tmpname, &st) < 0) {
534 warning("failed to stat %s: %s",
535 tmpname, strerror(errno));
536 } else if (!last_mtime) {
537 last_mtime = st.st_mtime;
540 utb.actime = st.st_atime;
541 utb.modtime = --last_mtime;
542 if (utime(tmpname, &utb) < 0)
543 warning("failed utime() on %s: %s",
544 tmpname, strerror(errno));
547 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
548 base_name, sha1_to_hex(sha1));
549 if (adjust_perm(idx_tmp_name, mode))
550 die("unable to make temporary index file readable: %s",
552 if (rename(idx_tmp_name, tmpname))
553 die("unable to rename temporary index file: %s",
558 puts(sha1_to_hex(sha1));
561 /* mark written objects as written to previous pack */
562 for (j = 0; j < nr_written; j++) {
563 written_list[j]->offset = (off_t)-1;
565 nr_remaining -= nr_written;
566 } while (nr_remaining && i < nr_objects);
569 stop_progress(&progress_state);
570 if (written != nr_result)
571 die("wrote %u objects while expecting %u", written, nr_result);
573 * We have scanned through [0 ... i). Since we have written
574 * the correct number of objects, the remaining [i ... nr_objects)
575 * items must be either already written (due to out-of-order delta base)
576 * or a preferred base. Count those which are neither and complain if any.
578 for (j = 0; i < nr_objects; i++) {
579 struct object_entry *e = objects + i;
580 j += !e->idx.offset && !e->preferred_base;
583 die("wrote %u objects as expected but %u unwritten", written, j);
586 static int locate_object_entry_hash(const unsigned char *sha1)
590 memcpy(&ui, sha1, sizeof(unsigned int));
591 i = ui % object_ix_hashsz;
592 while (0 < object_ix[i]) {
593 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
595 if (++i == object_ix_hashsz)
601 static struct object_entry *locate_object_entry(const unsigned char *sha1)
605 if (!object_ix_hashsz)
608 i = locate_object_entry_hash(sha1);
610 return &objects[object_ix[i]-1];
614 static void rehash_objects(void)
617 struct object_entry *oe;
619 object_ix_hashsz = nr_objects * 3;
620 if (object_ix_hashsz < 1024)
621 object_ix_hashsz = 1024;
622 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
623 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
624 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
625 int ix = locate_object_entry_hash(oe->idx.sha1);
629 object_ix[ix] = i + 1;
633 static unsigned name_hash(const char *name)
642 * This effectively just creates a sortable number from the
643 * last sixteen non-whitespace characters. Last characters
644 * count "most", so things that end in ".c" sort together.
646 while ((c = *name++) != 0) {
649 hash = (hash >> 2) + (c << 24);
654 static void setup_delta_attr_check(struct git_attr_check *check)
656 static struct git_attr *attr_delta;
659 attr_delta = git_attr("delta", 5);
661 check[0].attr = attr_delta;
664 static int no_try_delta(const char *path)
666 struct git_attr_check check[1];
668 setup_delta_attr_check(check);
669 if (git_checkattr(path, ARRAY_SIZE(check), check))
671 if (ATTR_FALSE(check->value))
676 static int add_object_entry(const unsigned char *sha1, enum object_type type,
677 const char *name, int exclude)
679 struct object_entry *entry;
680 struct packed_git *p, *found_pack = NULL;
681 off_t found_offset = 0;
683 unsigned hash = name_hash(name);
685 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
688 entry = objects + object_ix[ix] - 1;
689 if (!entry->preferred_base)
691 entry->preferred_base = 1;
696 for (p = packed_git; p; p = p->next) {
697 off_t offset = find_pack_entry_one(sha1, p);
700 found_offset = offset;
707 if (local && !p->pack_local)
712 if (nr_objects >= nr_alloc) {
713 nr_alloc = (nr_alloc + 1024) * 3 / 2;
714 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
717 entry = objects + nr_objects++;
718 memset(entry, 0, sizeof(*entry));
719 hashcpy(entry->idx.sha1, sha1);
724 entry->preferred_base = 1;
728 entry->in_pack = found_pack;
729 entry->in_pack_offset = found_offset;
732 if (object_ix_hashsz * 3 <= nr_objects * 4)
735 object_ix[-1 - ix] = nr_objects;
737 display_progress(progress_state, nr_objects);
739 if (name && no_try_delta(name))
740 entry->no_try_delta = 1;
745 struct pbase_tree_cache {
746 unsigned char sha1[20];
750 unsigned long tree_size;
753 static struct pbase_tree_cache *(pbase_tree_cache[256]);
754 static int pbase_tree_cache_ix(const unsigned char *sha1)
756 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
758 static int pbase_tree_cache_ix_incr(int ix)
760 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
763 static struct pbase_tree {
764 struct pbase_tree *next;
765 /* This is a phony "cache" entry; we are not
766 * going to evict it nor find it through _get()
767 * mechanism -- this is for the toplevel node that
768 * would almost always change with any commit.
770 struct pbase_tree_cache pcache;
773 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
775 struct pbase_tree_cache *ent, *nent;
778 enum object_type type;
780 int my_ix = pbase_tree_cache_ix(sha1);
781 int available_ix = -1;
783 /* pbase-tree-cache acts as a limited hashtable.
784 * your object will be found at your index or within a few
785 * slots after that slot if it is cached.
787 for (neigh = 0; neigh < 8; neigh++) {
788 ent = pbase_tree_cache[my_ix];
789 if (ent && !hashcmp(ent->sha1, sha1)) {
793 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
794 ((0 <= available_ix) &&
795 (!ent && pbase_tree_cache[available_ix])))
796 available_ix = my_ix;
799 my_ix = pbase_tree_cache_ix_incr(my_ix);
802 /* Did not find one. Either we got a bogus request or
803 * we need to read and perhaps cache.
805 data = read_sha1_file(sha1, &type, &size);
808 if (type != OBJ_TREE) {
813 /* We need to either cache or return a throwaway copy */
815 if (available_ix < 0)
818 ent = pbase_tree_cache[available_ix];
819 my_ix = available_ix;
823 nent = xmalloc(sizeof(*nent));
824 nent->temporary = (available_ix < 0);
827 /* evict and reuse */
828 free(ent->tree_data);
831 hashcpy(nent->sha1, sha1);
832 nent->tree_data = data;
833 nent->tree_size = size;
835 if (!nent->temporary)
836 pbase_tree_cache[my_ix] = nent;
840 static void pbase_tree_put(struct pbase_tree_cache *cache)
842 if (!cache->temporary) {
846 free(cache->tree_data);
850 static int name_cmp_len(const char *name)
853 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
858 static void add_pbase_object(struct tree_desc *tree,
861 const char *fullname)
863 struct name_entry entry;
866 while (tree_entry(tree,&entry)) {
867 if (S_ISGITLINK(entry.mode))
869 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
870 memcmp(name, entry.path, cmplen);
875 if (name[cmplen] != '/') {
876 add_object_entry(entry.sha1,
877 object_type(entry.mode),
881 if (S_ISDIR(entry.mode)) {
882 struct tree_desc sub;
883 struct pbase_tree_cache *tree;
884 const char *down = name+cmplen+1;
885 int downlen = name_cmp_len(down);
887 tree = pbase_tree_get(entry.sha1);
890 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
892 add_pbase_object(&sub, down, downlen, fullname);
893 pbase_tree_put(tree);
898 static unsigned *done_pbase_paths;
899 static int done_pbase_paths_num;
900 static int done_pbase_paths_alloc;
901 static int done_pbase_path_pos(unsigned hash)
904 int hi = done_pbase_paths_num;
906 int mi = (hi + lo) / 2;
907 if (done_pbase_paths[mi] == hash)
909 if (done_pbase_paths[mi] < hash)
917 static int check_pbase_path(unsigned hash)
919 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
923 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
924 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
925 done_pbase_paths = xrealloc(done_pbase_paths,
926 done_pbase_paths_alloc *
929 done_pbase_paths_num++;
930 if (pos < done_pbase_paths_num)
931 memmove(done_pbase_paths + pos + 1,
932 done_pbase_paths + pos,
933 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
934 done_pbase_paths[pos] = hash;
938 static void add_preferred_base_object(const char *name)
940 struct pbase_tree *it;
942 unsigned hash = name_hash(name);
944 if (!num_preferred_base || check_pbase_path(hash))
947 cmplen = name_cmp_len(name);
948 for (it = pbase_tree; it; it = it->next) {
950 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
953 struct tree_desc tree;
954 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
955 add_pbase_object(&tree, name, cmplen, name);
960 static void add_preferred_base(unsigned char *sha1)
962 struct pbase_tree *it;
965 unsigned char tree_sha1[20];
967 if (window <= num_preferred_base++)
970 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
974 for (it = pbase_tree; it; it = it->next) {
975 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
981 it = xcalloc(1, sizeof(*it));
982 it->next = pbase_tree;
985 hashcpy(it->pcache.sha1, tree_sha1);
986 it->pcache.tree_data = data;
987 it->pcache.tree_size = size;
990 static void check_object(struct object_entry *entry)
992 if (entry->in_pack) {
993 struct packed_git *p = entry->in_pack;
994 struct pack_window *w_curs = NULL;
995 const unsigned char *base_ref = NULL;
996 struct object_entry *base_entry;
997 unsigned long used, used_0;
1000 unsigned char *buf, c;
1002 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1005 * We want in_pack_type even if we do not reuse delta
1006 * since non-delta representations could still be reused.
1008 used = unpack_object_header_gently(buf, avail,
1009 &entry->in_pack_type,
1013 * Determine if this is a delta and if so whether we can
1014 * reuse it or not. Otherwise let's find out as cheaply as
1015 * possible what the actual type and size for this object is.
1017 switch (entry->in_pack_type) {
1019 /* Not a delta hence we've already got all we need. */
1020 entry->type = entry->in_pack_type;
1021 entry->in_pack_header_size = used;
1022 unuse_pack(&w_curs);
1025 if (reuse_delta && !entry->preferred_base)
1026 base_ref = use_pack(p, &w_curs,
1027 entry->in_pack_offset + used, NULL);
1028 entry->in_pack_header_size = used + 20;
1031 buf = use_pack(p, &w_curs,
1032 entry->in_pack_offset + used, NULL);
1038 if (!ofs || MSB(ofs, 7))
1039 die("delta base offset overflow in pack for %s",
1040 sha1_to_hex(entry->idx.sha1));
1042 ofs = (ofs << 7) + (c & 127);
1044 if (ofs >= entry->in_pack_offset)
1045 die("delta base offset out of bound for %s",
1046 sha1_to_hex(entry->idx.sha1));
1047 ofs = entry->in_pack_offset - ofs;
1048 if (reuse_delta && !entry->preferred_base) {
1049 struct revindex_entry *revidx;
1050 revidx = find_pack_revindex(p, ofs);
1051 base_ref = nth_packed_object_sha1(p, revidx->nr);
1053 entry->in_pack_header_size = used + used_0;
1057 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1059 * If base_ref was set above that means we wish to
1060 * reuse delta data, and we even found that base
1061 * in the list of objects we want to pack. Goodie!
1063 * Depth value does not matter - find_deltas() will
1064 * never consider reused delta as the base object to
1065 * deltify other objects against, in order to avoid
1068 entry->type = entry->in_pack_type;
1069 entry->delta = base_entry;
1070 entry->delta_sibling = base_entry->delta_child;
1071 base_entry->delta_child = entry;
1072 unuse_pack(&w_curs);
1078 * This must be a delta and we already know what the
1079 * final object type is. Let's extract the actual
1080 * object size from the delta header.
1082 entry->size = get_size_from_delta(p, &w_curs,
1083 entry->in_pack_offset + entry->in_pack_header_size);
1084 unuse_pack(&w_curs);
1089 * No choice but to fall back to the recursive delta walk
1090 * with sha1_object_info() to find about the object type
1093 unuse_pack(&w_curs);
1096 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1097 if (entry->type < 0)
1098 die("unable to get type of object %s",
1099 sha1_to_hex(entry->idx.sha1));
1102 static int pack_offset_sort(const void *_a, const void *_b)
1104 const struct object_entry *a = *(struct object_entry **)_a;
1105 const struct object_entry *b = *(struct object_entry **)_b;
1107 /* avoid filesystem trashing with loose objects */
1108 if (!a->in_pack && !b->in_pack)
1109 return hashcmp(a->idx.sha1, b->idx.sha1);
1111 if (a->in_pack < b->in_pack)
1113 if (a->in_pack > b->in_pack)
1115 return a->in_pack_offset < b->in_pack_offset ? -1 :
1116 (a->in_pack_offset > b->in_pack_offset);
1119 static void get_object_details(void)
1122 struct object_entry **sorted_by_offset;
1124 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1125 for (i = 0; i < nr_objects; i++)
1126 sorted_by_offset[i] = objects + i;
1127 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1129 for (i = 0; i < nr_objects; i++)
1130 check_object(sorted_by_offset[i]);
1132 free(sorted_by_offset);
1136 * We search for deltas in a list sorted by type, by filename hash, and then
1137 * by size, so that we see progressively smaller and smaller files.
1138 * That's because we prefer deltas to be from the bigger file
1139 * to the smaller -- deletes are potentially cheaper, but perhaps
1140 * more importantly, the bigger file is likely the more recent
1141 * one. The deepest deltas are therefore the oldest objects which are
1142 * less susceptible to be accessed often.
1144 static int type_size_sort(const void *_a, const void *_b)
1146 const struct object_entry *a = *(struct object_entry **)_a;
1147 const struct object_entry *b = *(struct object_entry **)_b;
1149 if (a->type > b->type)
1151 if (a->type < b->type)
1153 if (a->hash > b->hash)
1155 if (a->hash < b->hash)
1157 if (a->preferred_base > b->preferred_base)
1159 if (a->preferred_base < b->preferred_base)
1161 if (a->size > b->size)
1163 if (a->size < b->size)
1165 return a < b ? -1 : (a > b); /* newest first */
1169 struct object_entry *entry;
1171 struct delta_index *index;
1175 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1176 unsigned long delta_size)
1178 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1181 if (delta_size < cache_max_small_delta_size)
1184 /* cache delta, if objects are large enough compared to delta size */
1185 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1191 #ifdef THREADED_DELTA_SEARCH
1193 static pthread_mutex_t read_mutex = PTHREAD_MUTEX_INITIALIZER;
1194 #define read_lock() pthread_mutex_lock(&read_mutex)
1195 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1197 static pthread_mutex_t cache_mutex = PTHREAD_MUTEX_INITIALIZER;
1198 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1199 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1201 static pthread_mutex_t progress_mutex = PTHREAD_MUTEX_INITIALIZER;
1202 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1203 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1207 #define read_lock() (void)0
1208 #define read_unlock() (void)0
1209 #define cache_lock() (void)0
1210 #define cache_unlock() (void)0
1211 #define progress_lock() (void)0
1212 #define progress_unlock() (void)0
1216 static int try_delta(struct unpacked *trg, struct unpacked *src,
1217 unsigned max_depth, unsigned long *mem_usage)
1219 struct object_entry *trg_entry = trg->entry;
1220 struct object_entry *src_entry = src->entry;
1221 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1223 enum object_type type;
1226 /* Don't bother doing diffs between different types */
1227 if (trg_entry->type != src_entry->type)
1231 * We do not bother to try a delta that we discarded
1232 * on an earlier try, but only when reusing delta data.
1234 if (reuse_delta && trg_entry->in_pack &&
1235 trg_entry->in_pack == src_entry->in_pack &&
1236 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1237 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1240 /* Let's not bust the allowed depth. */
1241 if (src->depth >= max_depth)
1244 /* Now some size filtering heuristics. */
1245 trg_size = trg_entry->size;
1246 if (!trg_entry->delta) {
1247 max_size = trg_size/2 - 20;
1250 max_size = trg_entry->delta_size;
1251 ref_depth = trg->depth;
1253 max_size = max_size * (max_depth - src->depth) /
1254 (max_depth - ref_depth + 1);
1257 src_size = src_entry->size;
1258 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1259 if (sizediff >= max_size)
1261 if (trg_size < src_size / 32)
1264 /* Load data if not already done */
1267 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1270 die("object %s cannot be read",
1271 sha1_to_hex(trg_entry->idx.sha1));
1273 die("object %s inconsistent object length (%lu vs %lu)",
1274 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1279 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1282 die("object %s cannot be read",
1283 sha1_to_hex(src_entry->idx.sha1));
1285 die("object %s inconsistent object length (%lu vs %lu)",
1286 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1290 src->index = create_delta_index(src->data, src_size);
1292 static int warned = 0;
1294 warning("suboptimal pack - out of memory");
1297 *mem_usage += sizeof_delta_index(src->index);
1300 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1304 if (trg_entry->delta) {
1305 /* Prefer only shallower same-sized deltas. */
1306 if (delta_size == trg_entry->delta_size &&
1307 src->depth + 1 >= trg->depth) {
1314 * Handle memory allocation outside of the cache
1315 * accounting lock. Compiler will optimize the strangeness
1316 * away when THREADED_DELTA_SEARCH is not defined.
1318 free(trg_entry->delta_data);
1320 if (trg_entry->delta_data) {
1321 delta_cache_size -= trg_entry->delta_size;
1322 trg_entry->delta_data = NULL;
1324 if (delta_cacheable(src_size, trg_size, delta_size)) {
1325 delta_cache_size += delta_size;
1327 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1333 trg_entry->delta = src_entry;
1334 trg_entry->delta_size = delta_size;
1335 trg->depth = src->depth + 1;
1340 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1342 struct object_entry *child = me->delta_child;
1345 unsigned int c = check_delta_limit(child, n + 1);
1348 child = child->delta_sibling;
1353 static unsigned long free_unpacked(struct unpacked *n)
1355 unsigned long freed_mem = sizeof_delta_index(n->index);
1356 free_delta_index(n->index);
1359 freed_mem += n->entry->size;
1368 static void find_deltas(struct object_entry **list, unsigned *list_size,
1369 int window, int depth, unsigned *processed)
1371 uint32_t i, idx = 0, count = 0;
1372 unsigned int array_size = window * sizeof(struct unpacked);
1373 struct unpacked *array;
1374 unsigned long mem_usage = 0;
1376 array = xmalloc(array_size);
1377 memset(array, 0, array_size);
1380 struct object_entry *entry = *list++;
1381 struct unpacked *n = array + idx;
1382 int j, max_depth, best_base = -1;
1390 if (!entry->preferred_base) {
1392 display_progress(progress_state, *processed);
1396 mem_usage -= free_unpacked(n);
1399 while (window_memory_limit &&
1400 mem_usage > window_memory_limit &&
1402 uint32_t tail = (idx + window - count) % window;
1403 mem_usage -= free_unpacked(array + tail);
1407 /* We do not compute delta to *create* objects we are not
1410 if (entry->preferred_base)
1414 * If the current object is at pack edge, take the depth the
1415 * objects that depend on the current object into account
1416 * otherwise they would become too deep.
1419 if (entry->delta_child) {
1420 max_depth -= check_delta_limit(entry, 0);
1428 uint32_t other_idx = idx + j;
1430 if (other_idx >= window)
1431 other_idx -= window;
1432 m = array + other_idx;
1435 ret = try_delta(n, m, max_depth, &mem_usage);
1439 best_base = other_idx;
1443 * If we decided to cache the delta data, then it is best
1444 * to compress it right away. First because we have to do
1445 * it anyway, and doing it here while we're threaded will
1446 * save a lot of time in the non threaded write phase,
1447 * as well as allow for caching more deltas within
1448 * the same cache size limit.
1450 * But only if not writing to stdout, since in that case
1451 * the network is most likely throttling writes anyway,
1452 * and therefore it is best to go to the write phase ASAP
1453 * instead, as we can afford spending more time compressing
1454 * between writes at that moment.
1456 if (entry->delta_data && !pack_to_stdout) {
1457 entry->z_delta_size = do_compress(&entry->delta_data,
1460 delta_cache_size -= entry->delta_size;
1461 delta_cache_size += entry->z_delta_size;
1465 /* if we made n a delta, and if n is already at max
1466 * depth, leaving it in the window is pointless. we
1467 * should evict it first.
1469 if (entry->delta && max_depth <= n->depth)
1473 * Move the best delta base up in the window, after the
1474 * currently deltified object, to keep it longer. It will
1475 * be the first base object to be attempted next.
1478 struct unpacked swap = array[best_base];
1479 int dist = (window + idx - best_base) % window;
1480 int dst = best_base;
1482 int src = (dst + 1) % window;
1483 array[dst] = array[src];
1491 if (count + 1 < window)
1497 for (i = 0; i < window; ++i) {
1498 free_delta_index(array[i].index);
1499 free(array[i].data);
1504 #ifdef THREADED_DELTA_SEARCH
1507 * The main thread waits on the condition that (at least) one of the workers
1508 * has stopped working (which is indicated in the .working member of
1509 * struct thread_params).
1510 * When a work thread has completed its work, it sets .working to 0 and
1511 * signals the main thread and waits on the condition that .data_ready
1515 struct thread_params {
1517 struct object_entry **list;
1524 pthread_mutex_t mutex;
1525 pthread_cond_t cond;
1526 unsigned *processed;
1529 static pthread_cond_t progress_cond = PTHREAD_COND_INITIALIZER;
1531 static void *threaded_find_deltas(void *arg)
1533 struct thread_params *me = arg;
1535 while (me->remaining) {
1536 find_deltas(me->list, &me->remaining,
1537 me->window, me->depth, me->processed);
1541 pthread_cond_signal(&progress_cond);
1545 * We must not set ->data_ready before we wait on the
1546 * condition because the main thread may have set it to 1
1547 * before we get here. In order to be sure that new
1548 * work is available if we see 1 in ->data_ready, it
1549 * was initialized to 0 before this thread was spawned
1550 * and we reset it to 0 right away.
1552 pthread_mutex_lock(&me->mutex);
1553 while (!me->data_ready)
1554 pthread_cond_wait(&me->cond, &me->mutex);
1556 pthread_mutex_unlock(&me->mutex);
1558 /* leave ->working 1 so that this doesn't get more work assigned */
1562 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1563 int window, int depth, unsigned *processed)
1565 struct thread_params p[delta_search_threads];
1566 int i, ret, active_threads = 0;
1568 if (delta_search_threads <= 1) {
1569 find_deltas(list, &list_size, window, depth, processed);
1573 /* Partition the work amongst work threads. */
1574 for (i = 0; i < delta_search_threads; i++) {
1575 unsigned sub_size = list_size / (delta_search_threads - i);
1577 p[i].window = window;
1579 p[i].processed = processed;
1581 p[i].data_ready = 0;
1583 /* try to split chunks on "path" boundaries */
1584 while (sub_size && sub_size < list_size &&
1585 list[sub_size]->hash &&
1586 list[sub_size]->hash == list[sub_size-1]->hash)
1590 p[i].list_size = sub_size;
1591 p[i].remaining = sub_size;
1594 list_size -= sub_size;
1597 /* Start work threads. */
1598 for (i = 0; i < delta_search_threads; i++) {
1599 if (!p[i].list_size)
1601 pthread_mutex_init(&p[i].mutex, NULL);
1602 pthread_cond_init(&p[i].cond, NULL);
1603 ret = pthread_create(&p[i].thread, NULL,
1604 threaded_find_deltas, &p[i]);
1606 die("unable to create thread: %s", strerror(ret));
1611 * Now let's wait for work completion. Each time a thread is done
1612 * with its work, we steal half of the remaining work from the
1613 * thread with the largest number of unprocessed objects and give
1614 * it to that newly idle thread. This ensure good load balancing
1615 * until the remaining object list segments are simply too short
1616 * to be worth splitting anymore.
1618 while (active_threads) {
1619 struct thread_params *target = NULL;
1620 struct thread_params *victim = NULL;
1621 unsigned sub_size = 0;
1625 for (i = 0; !target && i < delta_search_threads; i++)
1630 pthread_cond_wait(&progress_cond, &progress_mutex);
1633 for (i = 0; i < delta_search_threads; i++)
1634 if (p[i].remaining > 2*window &&
1635 (!victim || victim->remaining < p[i].remaining))
1638 sub_size = victim->remaining / 2;
1639 list = victim->list + victim->list_size - sub_size;
1640 while (sub_size && list[0]->hash &&
1641 list[0]->hash == list[-1]->hash) {
1647 * It is possible for some "paths" to have
1648 * so many objects that no hash boundary
1649 * might be found. Let's just steal the
1650 * exact half in that case.
1652 sub_size = victim->remaining / 2;
1655 target->list = list;
1656 victim->list_size -= sub_size;
1657 victim->remaining -= sub_size;
1659 target->list_size = sub_size;
1660 target->remaining = sub_size;
1661 target->working = 1;
1664 pthread_mutex_lock(&target->mutex);
1665 target->data_ready = 1;
1666 pthread_cond_signal(&target->cond);
1667 pthread_mutex_unlock(&target->mutex);
1670 pthread_join(target->thread, NULL);
1671 pthread_cond_destroy(&target->cond);
1672 pthread_mutex_destroy(&target->mutex);
1679 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1682 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1684 unsigned char peeled[20];
1686 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1687 !peel_ref(path, peeled) && /* peelable? */
1688 !is_null_sha1(peeled) && /* annotated tag? */
1689 locate_object_entry(peeled)) /* object packed? */
1690 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1694 static void prepare_pack(int window, int depth)
1696 struct object_entry **delta_list;
1697 uint32_t i, n, nr_deltas;
1699 get_object_details();
1701 if (!nr_objects || !window || !depth)
1704 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1707 for (i = 0; i < nr_objects; i++) {
1708 struct object_entry *entry = objects + i;
1711 /* This happens if we decided to reuse existing
1712 * delta from a pack. "reuse_delta &&" is implied.
1716 if (entry->size < 50)
1719 if (entry->no_try_delta)
1722 if (!entry->preferred_base)
1725 delta_list[n++] = entry;
1728 if (nr_deltas && n > 1) {
1729 unsigned nr_done = 0;
1731 progress_state = start_progress("Compressing objects",
1733 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1734 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1735 stop_progress(&progress_state);
1736 if (nr_done != nr_deltas)
1737 die("inconsistency with delta count");
1742 static int git_pack_config(const char *k, const char *v, void *cb)
1744 if(!strcmp(k, "pack.window")) {
1745 window = git_config_int(k, v);
1748 if (!strcmp(k, "pack.windowmemory")) {
1749 window_memory_limit = git_config_ulong(k, v);
1752 if (!strcmp(k, "pack.depth")) {
1753 depth = git_config_int(k, v);
1756 if (!strcmp(k, "pack.compression")) {
1757 int level = git_config_int(k, v);
1759 level = Z_DEFAULT_COMPRESSION;
1760 else if (level < 0 || level > Z_BEST_COMPRESSION)
1761 die("bad pack compression level %d", level);
1762 pack_compression_level = level;
1763 pack_compression_seen = 1;
1766 if (!strcmp(k, "pack.deltacachesize")) {
1767 max_delta_cache_size = git_config_int(k, v);
1770 if (!strcmp(k, "pack.deltacachelimit")) {
1771 cache_max_small_delta_size = git_config_int(k, v);
1774 if (!strcmp(k, "pack.threads")) {
1775 delta_search_threads = git_config_int(k, v);
1776 if (delta_search_threads < 0)
1777 die("invalid number of threads specified (%d)",
1778 delta_search_threads);
1779 #ifndef THREADED_DELTA_SEARCH
1780 if (delta_search_threads != 1)
1781 warning("no threads support, ignoring %s", k);
1785 if (!strcmp(k, "pack.indexversion")) {
1786 pack_idx_default_version = git_config_int(k, v);
1787 if (pack_idx_default_version > 2)
1788 die("bad pack.indexversion=%d", pack_idx_default_version);
1791 if (!strcmp(k, "pack.packsizelimit")) {
1792 pack_size_limit_cfg = git_config_ulong(k, v);
1795 return git_default_config(k, v, cb);
1798 static void read_object_list_from_stdin(void)
1800 char line[40 + 1 + PATH_MAX + 2];
1801 unsigned char sha1[20];
1804 if (!fgets(line, sizeof(line), stdin)) {
1808 die("fgets returned NULL, not EOF, not error!");
1810 die("fgets: %s", strerror(errno));
1814 if (line[0] == '-') {
1815 if (get_sha1_hex(line+1, sha1))
1816 die("expected edge sha1, got garbage:\n %s",
1818 add_preferred_base(sha1);
1821 if (get_sha1_hex(line, sha1))
1822 die("expected sha1, got garbage:\n %s", line);
1824 add_preferred_base_object(line+41);
1825 add_object_entry(sha1, 0, line+41, 0);
1829 #define OBJECT_ADDED (1u<<20)
1831 static void show_commit(struct commit *commit)
1833 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1834 commit->object.flags |= OBJECT_ADDED;
1837 static void show_object(struct object_array_entry *p)
1839 add_preferred_base_object(p->name);
1840 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1841 p->item->flags |= OBJECT_ADDED;
1844 static void show_edge(struct commit *commit)
1846 add_preferred_base(commit->object.sha1);
1849 struct in_pack_object {
1851 struct object *object;
1857 struct in_pack_object *array;
1860 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1862 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1863 in_pack->array[in_pack->nr].object = object;
1868 * Compare the objects in the offset order, in order to emulate the
1869 * "git-rev-list --objects" output that produced the pack originally.
1871 static int ofscmp(const void *a_, const void *b_)
1873 struct in_pack_object *a = (struct in_pack_object *)a_;
1874 struct in_pack_object *b = (struct in_pack_object *)b_;
1876 if (a->offset < b->offset)
1878 else if (a->offset > b->offset)
1881 return hashcmp(a->object->sha1, b->object->sha1);
1884 static void add_objects_in_unpacked_packs(struct rev_info *revs)
1886 struct packed_git *p;
1887 struct in_pack in_pack;
1890 memset(&in_pack, 0, sizeof(in_pack));
1892 for (p = packed_git; p; p = p->next) {
1893 const unsigned char *sha1;
1896 for (i = 0; i < revs->num_ignore_packed; i++) {
1897 if (matches_pack_name(p, revs->ignore_packed[i]))
1900 if (revs->num_ignore_packed <= i)
1902 if (open_pack_index(p))
1903 die("cannot open pack index");
1905 ALLOC_GROW(in_pack.array,
1906 in_pack.nr + p->num_objects,
1909 for (i = 0; i < p->num_objects; i++) {
1910 sha1 = nth_packed_object_sha1(p, i);
1911 o = lookup_unknown_object(sha1);
1912 if (!(o->flags & OBJECT_ADDED))
1913 mark_in_pack_object(o, p, &in_pack);
1914 o->flags |= OBJECT_ADDED;
1919 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
1921 for (i = 0; i < in_pack.nr; i++) {
1922 struct object *o = in_pack.array[i].object;
1923 add_object_entry(o->sha1, o->type, "", 0);
1926 free(in_pack.array);
1929 static void loosen_unused_packed_objects(struct rev_info *revs)
1931 struct packed_git *p;
1933 const unsigned char *sha1;
1935 for (p = packed_git; p; p = p->next) {
1936 for (i = 0; i < revs->num_ignore_packed; i++) {
1937 if (matches_pack_name(p, revs->ignore_packed[i]))
1940 if (revs->num_ignore_packed <= i)
1943 if (open_pack_index(p))
1944 die("cannot open pack index");
1946 for (i = 0; i < p->num_objects; i++) {
1947 sha1 = nth_packed_object_sha1(p, i);
1948 if (!locate_object_entry(sha1))
1949 if (force_object_loose(sha1, p->mtime))
1950 die("unable to force loose object");
1955 static void get_object_list(int ac, const char **av)
1957 struct rev_info revs;
1961 init_revisions(&revs, NULL);
1962 save_commit_buffer = 0;
1963 setup_revisions(ac, av, &revs, NULL);
1965 while (fgets(line, sizeof(line), stdin) != NULL) {
1966 int len = strlen(line);
1967 if (len && line[len - 1] == '\n')
1972 if (!strcmp(line, "--not")) {
1973 flags ^= UNINTERESTING;
1976 die("not a rev '%s'", line);
1978 if (handle_revision_arg(line, &revs, flags, 1))
1979 die("bad revision '%s'", line);
1982 if (prepare_revision_walk(&revs))
1983 die("revision walk setup failed");
1984 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1985 traverse_commit_list(&revs, show_commit, show_object);
1987 if (keep_unreachable)
1988 add_objects_in_unpacked_packs(&revs);
1989 if (unpack_unreachable)
1990 loosen_unused_packed_objects(&revs);
1993 static int adjust_perm(const char *path, mode_t mode)
1995 if (chmod(path, mode))
1997 return adjust_shared_perm(path);
2000 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2002 int use_internal_rev_list = 0;
2006 int rp_ac_alloc = 64;
2009 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2011 rp_av[0] = "pack-objects";
2012 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2015 git_config(git_pack_config, NULL);
2016 if (!pack_compression_seen && core_compression_seen)
2017 pack_compression_level = core_compression_level;
2019 progress = isatty(2);
2020 for (i = 1; i < argc; i++) {
2021 const char *arg = argv[i];
2026 if (!strcmp("--non-empty", arg)) {
2030 if (!strcmp("--local", arg)) {
2034 if (!strcmp("--incremental", arg)) {
2038 if (!prefixcmp(arg, "--compression=")) {
2040 int level = strtoul(arg+14, &end, 0);
2041 if (!arg[14] || *end)
2044 level = Z_DEFAULT_COMPRESSION;
2045 else if (level < 0 || level > Z_BEST_COMPRESSION)
2046 die("bad pack compression level %d", level);
2047 pack_compression_level = level;
2050 if (!prefixcmp(arg, "--max-pack-size=")) {
2052 pack_size_limit_cfg = 0;
2053 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2054 if (!arg[16] || *end)
2058 if (!prefixcmp(arg, "--window=")) {
2060 window = strtoul(arg+9, &end, 0);
2061 if (!arg[9] || *end)
2065 if (!prefixcmp(arg, "--window-memory=")) {
2066 if (!git_parse_ulong(arg+16, &window_memory_limit))
2070 if (!prefixcmp(arg, "--threads=")) {
2072 delta_search_threads = strtoul(arg+10, &end, 0);
2073 if (!arg[10] || *end || delta_search_threads < 0)
2075 #ifndef THREADED_DELTA_SEARCH
2076 if (delta_search_threads != 1)
2077 warning("no threads support, "
2078 "ignoring %s", arg);
2082 if (!prefixcmp(arg, "--depth=")) {
2084 depth = strtoul(arg+8, &end, 0);
2085 if (!arg[8] || *end)
2089 if (!strcmp("--progress", arg)) {
2093 if (!strcmp("--all-progress", arg)) {
2097 if (!strcmp("-q", arg)) {
2101 if (!strcmp("--no-reuse-delta", arg)) {
2105 if (!strcmp("--no-reuse-object", arg)) {
2106 reuse_object = reuse_delta = 0;
2109 if (!strcmp("--delta-base-offset", arg)) {
2110 allow_ofs_delta = 1;
2113 if (!strcmp("--stdout", arg)) {
2117 if (!strcmp("--revs", arg)) {
2118 use_internal_rev_list = 1;
2121 if (!strcmp("--keep-unreachable", arg)) {
2122 keep_unreachable = 1;
2125 if (!strcmp("--unpack-unreachable", arg)) {
2126 unpack_unreachable = 1;
2129 if (!strcmp("--include-tag", arg)) {
2133 if (!strcmp("--unpacked", arg) ||
2134 !prefixcmp(arg, "--unpacked=") ||
2135 !strcmp("--reflog", arg) ||
2136 !strcmp("--all", arg)) {
2137 use_internal_rev_list = 1;
2138 if (rp_ac >= rp_ac_alloc - 1) {
2139 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2140 rp_av = xrealloc(rp_av,
2141 rp_ac_alloc * sizeof(*rp_av));
2143 rp_av[rp_ac++] = arg;
2146 if (!strcmp("--thin", arg)) {
2147 use_internal_rev_list = 1;
2149 rp_av[1] = "--objects-edge";
2152 if (!prefixcmp(arg, "--index-version=")) {
2154 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2155 if (pack_idx_default_version > 2)
2158 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2159 if (*c || pack_idx_off32_limit & 0x80000000)
2166 /* Traditionally "pack-objects [options] base extra" failed;
2167 * we would however want to take refs parameter that would
2168 * have been given to upstream rev-list ourselves, which means
2169 * we somehow want to say what the base name is. So the
2172 * pack-objects [options] base <refs...>
2174 * in other words, we would treat the first non-option as the
2175 * base_name and send everything else to the internal revision
2179 if (!pack_to_stdout)
2180 base_name = argv[i++];
2182 if (pack_to_stdout != !base_name)
2185 if (!pack_to_stdout && !pack_size_limit)
2186 pack_size_limit = pack_size_limit_cfg;
2188 if (pack_to_stdout && pack_size_limit)
2189 die("--max-pack-size cannot be used to build a pack for transfer.");
2191 if (!pack_to_stdout && thin)
2192 die("--thin cannot be used to build an indexable pack.");
2194 if (keep_unreachable && unpack_unreachable)
2195 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2197 #ifdef THREADED_DELTA_SEARCH
2198 if (!delta_search_threads) /* --threads=0 means autodetect */
2199 delta_search_threads = online_cpus();
2202 prepare_packed_git();
2205 progress_state = start_progress("Counting objects", 0);
2206 if (!use_internal_rev_list)
2207 read_object_list_from_stdin();
2209 rp_av[rp_ac] = NULL;
2210 get_object_list(rp_ac, rp_av);
2212 if (include_tag && nr_result)
2213 for_each_ref(add_ref_tag, NULL);
2214 stop_progress(&progress_state);
2216 if (non_empty && !nr_result)
2219 prepare_pack(window, depth);
2222 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
2223 written, written_delta, reused, reused_delta);