10 #include "csum-file.h"
11 #include "tree-walk.h"
15 static const char pack_usage
[] = "git-pack-objects [-q] [--no-reuse-delta] [--non-empty] [--local] [--incremental] [--window=N] [--depth=N] {--stdout | base-name} < object-list";
18 unsigned char sha1
[20];
19 unsigned long size
; /* uncompressed size */
20 unsigned long offset
; /* offset into the final pack file;
21 * nonzero if already written.
23 unsigned int depth
; /* delta depth */
24 unsigned int delta_limit
; /* base adjustment for in-pack delta */
25 unsigned int hash
; /* name hint hash */
26 enum object_type type
;
27 enum object_type in_pack_type
; /* could be delta */
28 unsigned long delta_size
; /* delta data size (uncompressed) */
29 struct object_entry
*delta
; /* delta base object */
30 struct packed_git
*in_pack
; /* already in pack */
31 unsigned int in_pack_offset
;
32 struct object_entry
*delta_child
; /* deltified objects who bases me */
33 struct object_entry
*delta_sibling
; /* other deltified objects who
34 * uses the same base as me
36 int preferred_base
; /* we do not pack this, but is encouraged to
37 * be used as the base objectto delta huge
43 * Objects we are going to pack are collected in objects array (dynamically
44 * expanded). nr_objects & nr_alloc controls this array. They are stored
45 * in the order we see -- typically rev-list --objects order that gives us
46 * nice "minimum seek" order.
48 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
49 * elements in the objects array. The former is used to build the pack
50 * index (lists object names in the ascending order to help offset lookup),
51 * and the latter is used to group similar things together by try_delta()
55 static unsigned char object_list_sha1
[20];
57 static int no_reuse_delta
;
59 static int incremental
;
60 static struct object_entry
**sorted_by_sha
, **sorted_by_type
;
61 static struct object_entry
*objects
;
62 static int nr_objects
, nr_alloc
, nr_result
;
63 static const char *base_name
;
64 static unsigned char pack_file_sha1
[20];
65 static int progress
= 1;
66 static volatile sig_atomic_t progress_update
;
67 static int window
= 10;
70 * The object names in objects array are hashed with this hashtable,
71 * to help looking up the entry by object name. Binary search from
72 * sorted_by_sha is also possible but this was easier to code and faster.
73 * This hashtable is built after all the objects are seen.
75 static int *object_ix
;
76 static int object_ix_hashsz
;
79 * Pack index for existing packs give us easy access to the offsets into
80 * corresponding pack file where each object's data starts, but the entries
81 * do not store the size of the compressed representation (uncompressed
82 * size is easily available by examining the pack entry header). We build
83 * a hashtable of existing packs (pack_revindex), and keep reverse index
84 * here -- pack index file is sorted by object name mapping to offset; this
85 * pack_revindex[].revindex array is an ordered list of offsets, so if you
86 * know the offset of an object, next offset is where its packed
87 * representation ends.
89 struct pack_revindex
{
91 unsigned long *revindex
;
92 } *pack_revindex
= NULL
;
93 static int pack_revindex_hashsz
;
99 static int written_delta
;
101 static int reused_delta
;
103 static int pack_revindex_ix(struct packed_git
*p
)
105 unsigned long ui
= (unsigned long)p
;
108 ui
= ui
^ (ui
>> 16); /* defeat structure alignment */
109 i
= (int)(ui
% pack_revindex_hashsz
);
110 while (pack_revindex
[i
].p
) {
111 if (pack_revindex
[i
].p
== p
)
113 if (++i
== pack_revindex_hashsz
)
119 static void prepare_pack_ix(void)
122 struct packed_git
*p
;
123 for (num
= 0, p
= packed_git
; p
; p
= p
->next
)
127 pack_revindex_hashsz
= num
* 11;
128 pack_revindex
= xcalloc(sizeof(*pack_revindex
), pack_revindex_hashsz
);
129 for (p
= packed_git
; p
; p
= p
->next
) {
130 num
= pack_revindex_ix(p
);
132 pack_revindex
[num
].p
= p
;
134 /* revindex elements are lazily initialized */
137 static int cmp_offset(const void *a_
, const void *b_
)
139 unsigned long a
= *(unsigned long *) a_
;
140 unsigned long b
= *(unsigned long *) b_
;
150 * Ordered list of offsets of objects in the pack.
152 static void prepare_pack_revindex(struct pack_revindex
*rix
)
154 struct packed_git
*p
= rix
->p
;
155 int num_ent
= num_packed_objects(p
);
157 void *index
= p
->index_base
+ 256;
159 rix
->revindex
= xmalloc(sizeof(unsigned long) * (num_ent
+ 1));
160 for (i
= 0; i
< num_ent
; i
++) {
161 unsigned int hl
= *((unsigned int *)((char *) index
+ 24*i
));
162 rix
->revindex
[i
] = ntohl(hl
);
164 /* This knows the pack format -- the 20-byte trailer
165 * follows immediately after the last object data.
167 rix
->revindex
[num_ent
] = p
->pack_size
- 20;
168 qsort(rix
->revindex
, num_ent
, sizeof(unsigned long), cmp_offset
);
171 static unsigned long find_packed_object_size(struct packed_git
*p
,
176 struct pack_revindex
*rix
;
177 unsigned long *revindex
;
178 num
= pack_revindex_ix(p
);
180 die("internal error: pack revindex uninitialized");
181 rix
= &pack_revindex
[num
];
183 prepare_pack_revindex(rix
);
184 revindex
= rix
->revindex
;
186 hi
= num_packed_objects(p
) + 1;
188 int mi
= (lo
+ hi
) / 2;
189 if (revindex
[mi
] == ofs
) {
190 return revindex
[mi
+1] - ofs
;
192 else if (ofs
< revindex
[mi
])
197 die("internal error: pack revindex corrupt");
200 static void *delta_against(void *buf
, unsigned long size
, struct object_entry
*entry
)
202 unsigned long othersize
, delta_size
;
204 void *otherbuf
= read_sha1_file(entry
->delta
->sha1
, type
, &othersize
);
208 die("unable to read %s", sha1_to_hex(entry
->delta
->sha1
));
209 delta_buf
= diff_delta(otherbuf
, othersize
,
210 buf
, size
, &delta_size
, 0);
211 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
212 die("delta size changed");
219 * The per-object header is a pretty dense thing, which is
220 * - first byte: low four bits are "size", then three bits of "type",
221 * and the high bit is "size continues".
222 * - each byte afterwards: low seven bits are size continuation,
223 * with the high bit being "size continues"
225 static int encode_header(enum object_type type
, unsigned long size
, unsigned char *hdr
)
230 if (type
< OBJ_COMMIT
|| type
> OBJ_DELTA
)
231 die("bad type %d", type
);
233 c
= (type
<< 4) | (size
& 15);
245 static unsigned long write_object(struct sha1file
*f
,
246 struct object_entry
*entry
)
251 unsigned char header
[10];
252 unsigned hdrlen
, datalen
;
253 enum object_type obj_type
;
256 if (entry
->preferred_base
)
259 obj_type
= entry
->type
;
260 if (! entry
->in_pack
)
261 to_reuse
= 0; /* can't reuse what we don't have */
262 else if (obj_type
== OBJ_DELTA
)
263 to_reuse
= 1; /* check_object() decided it for us */
264 else if (obj_type
!= entry
->in_pack_type
)
265 to_reuse
= 0; /* pack has delta which is unusable */
266 else if (entry
->delta
)
267 to_reuse
= 0; /* we want to pack afresh */
269 to_reuse
= 1; /* we have it in-pack undeltified,
270 * and we do not need to deltify it.
273 if (!entry
->in_pack
&& !entry
->delta
) {
275 unsigned long mapsize
;
276 map
= map_sha1_file(entry
->sha1
, &mapsize
);
277 if (map
&& !legacy_loose_object(map
)) {
278 /* We can copy straight into the pack file */
279 sha1write(f
, map
, mapsize
);
280 munmap(map
, mapsize
);
286 munmap(map
, mapsize
);
290 buf
= read_sha1_file(entry
->sha1
, type
, &size
);
292 die("unable to read %s", sha1_to_hex(entry
->sha1
));
293 if (size
!= entry
->size
)
294 die("object %s size inconsistency (%lu vs %lu)",
295 sha1_to_hex(entry
->sha1
), size
, entry
->size
);
297 buf
= delta_against(buf
, size
, entry
);
298 size
= entry
->delta_size
;
299 obj_type
= OBJ_DELTA
;
302 * The object header is a byte of 'type' followed by zero or
303 * more bytes of length. For deltas, the 20 bytes of delta
306 hdrlen
= encode_header(obj_type
, size
, header
);
307 sha1write(f
, header
, hdrlen
);
310 sha1write(f
, entry
->delta
, 20);
313 datalen
= sha1write_compressed(f
, buf
, size
);
317 struct packed_git
*p
= entry
->in_pack
;
320 datalen
= find_packed_object_size(p
, entry
->in_pack_offset
);
321 buf
= (char *) p
->pack_base
+ entry
->in_pack_offset
;
322 sha1write(f
, buf
, datalen
);
324 hdrlen
= 0; /* not really */
325 if (obj_type
== OBJ_DELTA
)
329 if (obj_type
== OBJ_DELTA
)
332 return hdrlen
+ datalen
;
335 static unsigned long write_one(struct sha1file
*f
,
336 struct object_entry
*e
,
337 unsigned long offset
)
340 /* offset starts from header size and cannot be zero
341 * if it is written already.
345 offset
+= write_object(f
, e
);
346 /* if we are deltified, write out its base object. */
348 offset
= write_one(f
, e
->delta
, offset
);
352 static void write_pack_file(void)
356 unsigned long offset
;
357 struct pack_header hdr
;
358 unsigned last_percent
= 999;
362 f
= sha1fd(1, "<stdout>");
364 f
= sha1create("%s-%s.%s", base_name
,
365 sha1_to_hex(object_list_sha1
), "pack");
366 do_progress
= progress
;
369 fprintf(stderr
, "Writing %d objects.\n", nr_result
);
371 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
372 hdr
.hdr_version
= htonl(PACK_VERSION
);
373 hdr
.hdr_entries
= htonl(nr_result
);
374 sha1write(f
, &hdr
, sizeof(hdr
));
375 offset
= sizeof(hdr
);
378 for (i
= 0; i
< nr_objects
; i
++) {
379 offset
= write_one(f
, objects
+ i
, offset
);
381 unsigned percent
= written
* 100 / nr_result
;
382 if (progress_update
|| percent
!= last_percent
) {
383 fprintf(stderr
, "%4u%% (%u/%u) done\r",
384 percent
, written
, nr_result
);
386 last_percent
= percent
;
393 sha1close(f
, pack_file_sha1
, 1);
396 static void write_index_file(void)
399 struct sha1file
*f
= sha1create("%s-%s.%s", base_name
,
400 sha1_to_hex(object_list_sha1
), "idx");
401 struct object_entry
**list
= sorted_by_sha
;
402 struct object_entry
**last
= list
+ nr_result
;
403 unsigned int array
[256];
406 * Write the first-level table (the list is sorted,
407 * but we use a 256-entry lookup to be able to avoid
408 * having to do eight extra binary search iterations).
410 for (i
= 0; i
< 256; i
++) {
411 struct object_entry
**next
= list
;
412 while (next
< last
) {
413 struct object_entry
*entry
= *next
;
414 if (entry
->sha1
[0] != i
)
418 array
[i
] = htonl(next
- sorted_by_sha
);
421 sha1write(f
, array
, 256 * sizeof(int));
424 * Write the actual SHA1 entries..
426 list
= sorted_by_sha
;
427 for (i
= 0; i
< nr_result
; i
++) {
428 struct object_entry
*entry
= *list
++;
429 unsigned int offset
= htonl(entry
->offset
);
430 sha1write(f
, &offset
, 4);
431 sha1write(f
, entry
->sha1
, 20);
433 sha1write(f
, pack_file_sha1
, 20);
434 sha1close(f
, NULL
, 1);
437 static int locate_object_entry_hash(const unsigned char *sha1
)
441 memcpy(&ui
, sha1
, sizeof(unsigned int));
442 i
= ui
% object_ix_hashsz
;
443 while (0 < object_ix
[i
]) {
444 if (!memcmp(sha1
, objects
[object_ix
[i
]-1].sha1
, 20))
446 if (++i
== object_ix_hashsz
)
452 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
456 if (!object_ix_hashsz
)
459 i
= locate_object_entry_hash(sha1
);
461 return &objects
[object_ix
[i
]-1];
465 static void rehash_objects(void)
468 struct object_entry
*oe
;
470 object_ix_hashsz
= nr_objects
* 3;
471 if (object_ix_hashsz
< 1024)
472 object_ix_hashsz
= 1024;
473 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
474 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
475 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
476 int ix
= locate_object_entry_hash(oe
->sha1
);
480 object_ix
[ix
] = i
+ 1;
484 static unsigned name_hash(const char *name
)
490 * This effectively just creates a sortable number from the
491 * last sixteen non-whitespace characters. Last characters
492 * count "most", so things that end in ".c" sort together.
494 while ((c
= *name
++) != 0) {
497 hash
= (hash
>> 2) + (c
<< 24);
502 static int add_object_entry(const unsigned char *sha1
, unsigned hash
, int exclude
)
504 unsigned int idx
= nr_objects
;
505 struct object_entry
*entry
;
506 struct packed_git
*p
;
507 unsigned int found_offset
= 0;
508 struct packed_git
*found_pack
= NULL
;
512 for (p
= packed_git
; p
; p
= p
->next
) {
514 if (find_pack_entry_one(sha1
, &e
, p
)) {
517 if (local
&& !p
->pack_local
)
520 found_offset
= e
.offset
;
526 if ((entry
= locate_object_entry(sha1
)) != NULL
)
529 if (idx
>= nr_alloc
) {
530 unsigned int needed
= (idx
+ 1024) * 3 / 2;
531 objects
= xrealloc(objects
, needed
* sizeof(*entry
));
534 entry
= objects
+ idx
;
535 nr_objects
= idx
+ 1;
536 memset(entry
, 0, sizeof(*entry
));
537 memcpy(entry
->sha1
, sha1
, 20);
540 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
543 ix
= locate_object_entry_hash(entry
->sha1
);
545 die("internal error in object hashing.");
546 object_ix
[-1 - ix
] = idx
+ 1;
551 if (progress_update
) {
552 fprintf(stderr
, "Counting objects...%d\r", nr_objects
);
556 entry
->preferred_base
= 1;
559 entry
->in_pack
= found_pack
;
560 entry
->in_pack_offset
= found_offset
;
566 struct pbase_tree_cache
{
567 unsigned char sha1
[20];
571 unsigned long tree_size
;
574 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
575 static int pbase_tree_cache_ix(const unsigned char *sha1
)
577 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
579 static int pbase_tree_cache_ix_incr(int ix
)
581 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
584 static struct pbase_tree
{
585 struct pbase_tree
*next
;
586 /* This is a phony "cache" entry; we are not
587 * going to evict it nor find it through _get()
588 * mechanism -- this is for the toplevel node that
589 * would almost always change with any commit.
591 struct pbase_tree_cache pcache
;
594 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
596 struct pbase_tree_cache
*ent
, *nent
;
601 int my_ix
= pbase_tree_cache_ix(sha1
);
602 int available_ix
= -1;
604 /* pbase-tree-cache acts as a limited hashtable.
605 * your object will be found at your index or within a few
606 * slots after that slot if it is cached.
608 for (neigh
= 0; neigh
< 8; neigh
++) {
609 ent
= pbase_tree_cache
[my_ix
];
610 if (ent
&& !memcmp(ent
->sha1
, sha1
, 20)) {
614 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
615 ((0 <= available_ix
) &&
616 (!ent
&& pbase_tree_cache
[available_ix
])))
617 available_ix
= my_ix
;
620 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
623 /* Did not find one. Either we got a bogus request or
624 * we need to read and perhaps cache.
626 data
= read_sha1_file(sha1
, type
, &size
);
629 if (strcmp(type
, tree_type
)) {
634 /* We need to either cache or return a throwaway copy */
636 if (available_ix
< 0)
639 ent
= pbase_tree_cache
[available_ix
];
640 my_ix
= available_ix
;
644 nent
= xmalloc(sizeof(*nent
));
645 nent
->temporary
= (available_ix
< 0);
648 /* evict and reuse */
649 free(ent
->tree_data
);
652 memcpy(nent
->sha1
, sha1
, 20);
653 nent
->tree_data
= data
;
654 nent
->tree_size
= size
;
656 if (!nent
->temporary
)
657 pbase_tree_cache
[my_ix
] = nent
;
661 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
663 if (!cache
->temporary
) {
667 free(cache
->tree_data
);
671 static int name_cmp_len(const char *name
)
674 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
679 static void add_pbase_object(struct tree_desc
*tree
,
682 const char *fullname
)
684 struct name_entry entry
;
686 while (tree_entry(tree
,&entry
)) {
690 if (entry
.pathlen
!= cmplen
||
691 memcmp(entry
.path
, name
, cmplen
) ||
692 !has_sha1_file(entry
.sha1
) ||
693 sha1_object_info(entry
.sha1
, type
, &size
))
695 if (name
[cmplen
] != '/') {
696 unsigned hash
= name_hash(fullname
);
697 add_object_entry(entry
.sha1
, hash
, 1);
700 if (!strcmp(type
, tree_type
)) {
701 struct tree_desc sub
;
702 struct pbase_tree_cache
*tree
;
703 const char *down
= name
+cmplen
+1;
704 int downlen
= name_cmp_len(down
);
706 tree
= pbase_tree_get(entry
.sha1
);
709 sub
.buf
= tree
->tree_data
;
710 sub
.size
= tree
->tree_size
;
712 add_pbase_object(&sub
, down
, downlen
, fullname
);
713 pbase_tree_put(tree
);
718 static unsigned *done_pbase_paths
;
719 static int done_pbase_paths_num
;
720 static int done_pbase_paths_alloc
;
721 static int done_pbase_path_pos(unsigned hash
)
724 int hi
= done_pbase_paths_num
;
726 int mi
= (hi
+ lo
) / 2;
727 if (done_pbase_paths
[mi
] == hash
)
729 if (done_pbase_paths
[mi
] < hash
)
737 static int check_pbase_path(unsigned hash
)
739 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
743 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
744 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
745 done_pbase_paths
= xrealloc(done_pbase_paths
,
746 done_pbase_paths_alloc
*
749 done_pbase_paths_num
++;
750 if (pos
< done_pbase_paths_num
)
751 memmove(done_pbase_paths
+ pos
+ 1,
752 done_pbase_paths
+ pos
,
753 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
754 done_pbase_paths
[pos
] = hash
;
758 static void add_preferred_base_object(char *name
, unsigned hash
)
760 struct pbase_tree
*it
;
761 int cmplen
= name_cmp_len(name
);
763 if (check_pbase_path(hash
))
766 for (it
= pbase_tree
; it
; it
= it
->next
) {
768 hash
= name_hash("");
769 add_object_entry(it
->pcache
.sha1
, hash
, 1);
772 struct tree_desc tree
;
773 tree
.buf
= it
->pcache
.tree_data
;
774 tree
.size
= it
->pcache
.tree_size
;
775 add_pbase_object(&tree
, name
, cmplen
, name
);
780 static void add_preferred_base(unsigned char *sha1
)
782 struct pbase_tree
*it
;
785 unsigned char tree_sha1
[20];
787 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
791 for (it
= pbase_tree
; it
; it
= it
->next
) {
792 if (!memcmp(it
->pcache
.sha1
, tree_sha1
, 20)) {
798 it
= xcalloc(1, sizeof(*it
));
799 it
->next
= pbase_tree
;
802 memcpy(it
->pcache
.sha1
, tree_sha1
, 20);
803 it
->pcache
.tree_data
= data
;
804 it
->pcache
.tree_size
= size
;
807 static void check_object(struct object_entry
*entry
)
811 if (entry
->in_pack
&& !entry
->preferred_base
) {
812 unsigned char base
[20];
814 struct object_entry
*base_entry
;
816 /* We want in_pack_type even if we do not reuse delta.
817 * There is no point not reusing non-delta representations.
819 check_reuse_pack_delta(entry
->in_pack
,
820 entry
->in_pack_offset
,
822 &entry
->in_pack_type
);
824 /* Check if it is delta, and the base is also an object
825 * we are going to pack. If so we will reuse the existing
828 if (!no_reuse_delta
&&
829 entry
->in_pack_type
== OBJ_DELTA
&&
830 (base_entry
= locate_object_entry(base
)) &&
831 (!base_entry
->preferred_base
)) {
833 /* Depth value does not matter - find_deltas()
834 * will never consider reused delta as the
835 * base object to deltify other objects
836 * against, in order to avoid circular deltas.
839 /* uncompressed size of the delta data */
840 entry
->size
= entry
->delta_size
= size
;
841 entry
->delta
= base_entry
;
842 entry
->type
= OBJ_DELTA
;
844 entry
->delta_sibling
= base_entry
->delta_child
;
845 base_entry
->delta_child
= entry
;
849 /* Otherwise we would do the usual */
852 if (sha1_object_info(entry
->sha1
, type
, &entry
->size
))
853 die("unable to get type of object %s",
854 sha1_to_hex(entry
->sha1
));
856 if (!strcmp(type
, commit_type
)) {
857 entry
->type
= OBJ_COMMIT
;
858 } else if (!strcmp(type
, tree_type
)) {
859 entry
->type
= OBJ_TREE
;
860 } else if (!strcmp(type
, blob_type
)) {
861 entry
->type
= OBJ_BLOB
;
862 } else if (!strcmp(type
, tag_type
)) {
863 entry
->type
= OBJ_TAG
;
865 die("unable to pack object %s of type %s",
866 sha1_to_hex(entry
->sha1
), type
);
869 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
871 struct object_entry
*child
= me
->delta_child
;
874 unsigned int c
= check_delta_limit(child
, n
+ 1);
877 child
= child
->delta_sibling
;
882 static void get_object_details(void)
885 struct object_entry
*entry
;
888 for (i
= 0, entry
= objects
; i
< nr_objects
; i
++, entry
++)
891 if (nr_objects
== nr_result
) {
893 * Depth of objects that depend on the entry -- this
894 * is subtracted from depth-max to break too deep
895 * delta chain because of delta data reusing.
896 * However, we loosen this restriction when we know we
897 * are creating a thin pack -- it will have to be
898 * expanded on the other end anyway, so do not
899 * artificially cut the delta chain and let it go as
902 for (i
= 0, entry
= objects
; i
< nr_objects
; i
++, entry
++)
903 if (!entry
->delta
&& entry
->delta_child
)
905 check_delta_limit(entry
, 1);
909 typedef int (*entry_sort_t
)(const struct object_entry
*, const struct object_entry
*);
911 static entry_sort_t current_sort
;
913 static int sort_comparator(const void *_a
, const void *_b
)
915 struct object_entry
*a
= *(struct object_entry
**)_a
;
916 struct object_entry
*b
= *(struct object_entry
**)_b
;
917 return current_sort(a
,b
);
920 static struct object_entry
**create_sorted_list(entry_sort_t sort
)
922 struct object_entry
**list
= xmalloc(nr_objects
* sizeof(struct object_entry
*));
925 for (i
= 0; i
< nr_objects
; i
++)
926 list
[i
] = objects
+ i
;
928 qsort(list
, nr_objects
, sizeof(struct object_entry
*), sort_comparator
);
932 static int sha1_sort(const struct object_entry
*a
, const struct object_entry
*b
)
934 return memcmp(a
->sha1
, b
->sha1
, 20);
937 static struct object_entry
**create_final_object_list(void)
939 struct object_entry
**list
;
942 for (i
= nr_result
= 0; i
< nr_objects
; i
++)
943 if (!objects
[i
].preferred_base
)
945 list
= xmalloc(nr_result
* sizeof(struct object_entry
*));
946 for (i
= j
= 0; i
< nr_objects
; i
++) {
947 if (!objects
[i
].preferred_base
)
948 list
[j
++] = objects
+ i
;
950 current_sort
= sha1_sort
;
951 qsort(list
, nr_result
, sizeof(struct object_entry
*), sort_comparator
);
955 static int type_size_sort(const struct object_entry
*a
, const struct object_entry
*b
)
957 if (a
->type
< b
->type
)
959 if (a
->type
> b
->type
)
961 if (a
->hash
< b
->hash
)
963 if (a
->hash
> b
->hash
)
965 if (a
->preferred_base
< b
->preferred_base
)
967 if (a
->preferred_base
> b
->preferred_base
)
969 if (a
->size
< b
->size
)
971 if (a
->size
> b
->size
)
973 return a
< b
? -1 : (a
> b
);
977 struct object_entry
*entry
;
979 struct delta_index
*index
;
983 * We search for deltas _backwards_ in a list sorted by type and
984 * by size, so that we see progressively smaller and smaller files.
985 * That's because we prefer deltas to be from the bigger file
986 * to the smaller - deletes are potentially cheaper, but perhaps
987 * more importantly, the bigger file is likely the more recent
990 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
993 struct object_entry
*trg_entry
= trg
->entry
;
994 struct object_entry
*src_entry
= src
->entry
;
995 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
999 /* Don't bother doing diffs between different types */
1000 if (trg_entry
->type
!= src_entry
->type
)
1003 /* We do not compute delta to *create* objects we are not
1006 if (trg_entry
->preferred_base
)
1010 * We do not bother to try a delta that we discarded
1011 * on an earlier try, but only when reusing delta data.
1013 if (!no_reuse_delta
&& trg_entry
->in_pack
&&
1014 trg_entry
->in_pack
== src_entry
->in_pack
)
1018 * If the current object is at pack edge, take the depth the
1019 * objects that depend on the current object into account --
1020 * otherwise they would become too deep.
1022 if (trg_entry
->delta_child
) {
1023 if (max_depth
<= trg_entry
->delta_limit
)
1025 max_depth
-= trg_entry
->delta_limit
;
1027 if (src_entry
->depth
>= max_depth
)
1030 /* Now some size filtering heuristics. */
1031 trg_size
= trg_entry
->size
;
1032 max_size
= trg_size
/2 - 20;
1033 max_size
= max_size
* (max_depth
- src_entry
->depth
) / max_depth
;
1036 if (trg_entry
->delta
&& trg_entry
->delta_size
<= max_size
)
1037 max_size
= trg_entry
->delta_size
-1;
1038 src_size
= src_entry
->size
;
1039 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1040 if (sizediff
>= max_size
)
1043 /* Load data if not already done */
1045 trg
->data
= read_sha1_file(trg_entry
->sha1
, type
, &sz
);
1047 die("object %s inconsistent object length (%lu vs %lu)",
1048 sha1_to_hex(trg_entry
->sha1
), sz
, trg_size
);
1051 src
->data
= read_sha1_file(src_entry
->sha1
, type
, &sz
);
1053 die("object %s inconsistent object length (%lu vs %lu)",
1054 sha1_to_hex(src_entry
->sha1
), sz
, src_size
);
1057 src
->index
= create_delta_index(src
->data
, src_size
);
1059 die("out of memory");
1062 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1066 trg_entry
->delta
= src_entry
;
1067 trg_entry
->delta_size
= delta_size
;
1068 trg_entry
->depth
= src_entry
->depth
+ 1;
1073 static void progress_interval(int signum
)
1075 progress_update
= 1;
1078 static void find_deltas(struct object_entry
**list
, int window
, int depth
)
1081 unsigned int array_size
= window
* sizeof(struct unpacked
);
1082 struct unpacked
*array
= xmalloc(array_size
);
1083 unsigned processed
= 0;
1084 unsigned last_percent
= 999;
1086 memset(array
, 0, array_size
);
1090 fprintf(stderr
, "Deltifying %d objects.\n", nr_result
);
1093 struct object_entry
*entry
= list
[i
];
1094 struct unpacked
*n
= array
+ idx
;
1097 if (!entry
->preferred_base
)
1101 unsigned percent
= processed
* 100 / nr_result
;
1102 if (percent
!= last_percent
|| progress_update
) {
1103 fprintf(stderr
, "%4u%% (%u/%u) done\r",
1104 percent
, processed
, nr_result
);
1105 progress_update
= 0;
1106 last_percent
= percent
;
1111 /* This happens if we decided to reuse existing
1112 * delta from a pack. "!no_reuse_delta &&" is implied.
1116 if (entry
->size
< 50)
1118 free_delta_index(n
->index
);
1126 unsigned int other_idx
= idx
+ j
;
1128 if (other_idx
>= window
)
1129 other_idx
-= window
;
1130 m
= array
+ other_idx
;
1133 if (try_delta(n
, m
, depth
) < 0)
1136 /* if we made n a delta, and if n is already at max
1137 * depth, leaving it in the window is pointless. we
1138 * should evict it first.
1140 if (entry
->delta
&& depth
<= entry
->depth
)
1149 fputc('\n', stderr
);
1151 for (i
= 0; i
< window
; ++i
) {
1152 free_delta_index(array
[i
].index
);
1153 free(array
[i
].data
);
1158 static void prepare_pack(int window
, int depth
)
1160 get_object_details();
1161 sorted_by_type
= create_sorted_list(type_size_sort
);
1162 if (window
&& depth
)
1163 find_deltas(sorted_by_type
, window
+1, depth
);
1166 static int reuse_cached_pack(unsigned char *sha1
, int pack_to_stdout
)
1168 static const char cache
[] = "pack-cache/pack-%s.%s";
1169 char *cached_pack
, *cached_idx
;
1170 int ifd
, ofd
, ifd_ix
= -1;
1172 cached_pack
= git_path(cache
, sha1_to_hex(sha1
), "pack");
1173 ifd
= open(cached_pack
, O_RDONLY
);
1177 if (!pack_to_stdout
) {
1178 cached_idx
= git_path(cache
, sha1_to_hex(sha1
), "idx");
1179 ifd_ix
= open(cached_idx
, O_RDONLY
);
1187 fprintf(stderr
, "Reusing %d objects pack %s\n", nr_objects
,
1190 if (pack_to_stdout
) {
1191 if (copy_fd(ifd
, 1))
1196 char name
[PATH_MAX
];
1197 snprintf(name
, sizeof(name
),
1198 "%s-%s.%s", base_name
, sha1_to_hex(sha1
), "pack");
1199 ofd
= open(name
, O_CREAT
| O_EXCL
| O_WRONLY
, 0666);
1201 die("unable to open %s (%s)", name
, strerror(errno
));
1202 if (copy_fd(ifd
, ofd
))
1206 snprintf(name
, sizeof(name
),
1207 "%s-%s.%s", base_name
, sha1_to_hex(sha1
), "idx");
1208 ofd
= open(name
, O_CREAT
| O_EXCL
| O_WRONLY
, 0666);
1210 die("unable to open %s (%s)", name
, strerror(errno
));
1211 if (copy_fd(ifd_ix
, ofd
))
1214 puts(sha1_to_hex(sha1
));
1220 static void setup_progress_signal(void)
1222 struct sigaction sa
;
1225 memset(&sa
, 0, sizeof(sa
));
1226 sa
.sa_handler
= progress_interval
;
1227 sigemptyset(&sa
.sa_mask
);
1228 sa
.sa_flags
= SA_RESTART
;
1229 sigaction(SIGALRM
, &sa
, NULL
);
1231 v
.it_interval
.tv_sec
= 1;
1232 v
.it_interval
.tv_usec
= 0;
1233 v
.it_value
= v
.it_interval
;
1234 setitimer(ITIMER_REAL
, &v
, NULL
);
1237 static int git_pack_config(const char *k
, const char *v
)
1239 if(!strcmp(k
, "pack.window")) {
1240 window
= git_config_int(k
, v
);
1243 return git_default_config(k
, v
);
1246 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
1249 char line
[40 + 1 + PATH_MAX
+ 2];
1250 int depth
= 10, pack_to_stdout
= 0;
1251 struct object_entry
**list
;
1252 int num_preferred_base
= 0;
1255 git_config(git_pack_config
);
1257 progress
= isatty(2);
1258 for (i
= 1; i
< argc
; i
++) {
1259 const char *arg
= argv
[i
];
1262 if (!strcmp("--non-empty", arg
)) {
1266 if (!strcmp("--local", arg
)) {
1270 if (!strcmp("--progress", arg
)) {
1274 if (!strcmp("--incremental", arg
)) {
1278 if (!strncmp("--window=", arg
, 9)) {
1280 window
= strtoul(arg
+9, &end
, 0);
1281 if (!arg
[9] || *end
)
1285 if (!strncmp("--depth=", arg
, 8)) {
1287 depth
= strtoul(arg
+8, &end
, 0);
1288 if (!arg
[8] || *end
)
1292 if (!strcmp("--progress", arg
)) {
1296 if (!strcmp("-q", arg
)) {
1300 if (!strcmp("--no-reuse-delta", arg
)) {
1304 if (!strcmp("--stdout", arg
)) {
1315 if (pack_to_stdout
!= !base_name
)
1318 prepare_packed_git();
1321 fprintf(stderr
, "Generating pack...\n");
1322 setup_progress_signal();
1326 unsigned char sha1
[20];
1329 if (!fgets(line
, sizeof(line
), stdin
)) {
1333 die("fgets returned NULL, not EOF, not error!");
1335 die("fgets: %s", strerror(errno
));
1340 if (line
[0] == '-') {
1341 if (get_sha1_hex(line
+1, sha1
))
1342 die("expected edge sha1, got garbage:\n %s",
1344 if (num_preferred_base
++ < window
)
1345 add_preferred_base(sha1
);
1348 if (get_sha1_hex(line
, sha1
))
1349 die("expected sha1, got garbage:\n %s", line
);
1350 hash
= name_hash(line
+41);
1351 add_preferred_base_object(line
+41, hash
);
1352 add_object_entry(sha1
, hash
, 0);
1355 fprintf(stderr
, "Done counting %d objects.\n", nr_objects
);
1356 sorted_by_sha
= create_final_object_list();
1357 if (non_empty
&& !nr_result
)
1361 list
= sorted_by_sha
;
1362 for (i
= 0; i
< nr_result
; i
++) {
1363 struct object_entry
*entry
= *list
++;
1364 SHA1_Update(&ctx
, entry
->sha1
, 20);
1366 SHA1_Final(object_list_sha1
, &ctx
);
1367 if (progress
&& (nr_objects
!= nr_result
))
1368 fprintf(stderr
, "Result has %d objects.\n", nr_result
);
1370 if (reuse_cached_pack(object_list_sha1
, pack_to_stdout
))
1374 prepare_pack(window
, depth
);
1375 if (progress
&& pack_to_stdout
) {
1376 /* the other end usually displays progress itself */
1377 struct itimerval v
= {{0,},};
1378 setitimer(ITIMER_REAL
, &v
, NULL
);
1379 signal(SIGALRM
, SIG_IGN
);
1380 progress_update
= 0;
1383 if (!pack_to_stdout
) {
1385 puts(sha1_to_hex(object_list_sha1
));
1389 fprintf(stderr
, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1390 nr_result
, written
, written_delta
, reused
, reused_delta
);