10 #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
; /* delitified 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 colected 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];
56 static int non_empty
= 0;
57 static int no_reuse_delta
= 0;
59 static int incremental
= 0;
60 static struct object_entry
**sorted_by_sha
, **sorted_by_type
;
61 static struct object_entry
*objects
= NULL
;
62 static int nr_objects
= 0, nr_alloc
= 0, nr_result
= 0;
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
= 0;
69 * The object names in objects array are hashed with this hashtable,
70 * to help looking up the entry by object name. Binary search from
71 * sorted_by_sha is also possible but this was easier to code and faster.
72 * This hashtable is built after all the objects are seen.
74 static int *object_ix
= NULL
;
75 static int object_ix_hashsz
= 0;
78 * Pack index for existing packs give us easy access to the offsets into
79 * corresponding pack file where each object's data starts, but the entries
80 * do not store the size of the compressed representation (uncompressed
81 * size is easily available by examining the pack entry header). We build
82 * a hashtable of existing packs (pack_revindex), and keep reverse index
83 * here -- pack index file is sorted by object name mapping to offset; this
84 * pack_revindex[].revindex array is an ordered list of offsets, so if you
85 * know the offset of an object, next offset is where its packed
86 * representation ends.
88 struct pack_revindex
{
90 unsigned long *revindex
;
91 } *pack_revindex
= NULL
;
92 static int pack_revindex_hashsz
= 0;
97 static int written
= 0;
98 static int written_delta
= 0;
99 static int reused
= 0;
100 static int reused_delta
= 0;
102 static int pack_revindex_ix(struct packed_git
*p
)
104 unsigned long ui
= (unsigned long)p
;
107 ui
= ui
^ (ui
>> 16); /* defeat structure alignment */
108 i
= (int)(ui
% pack_revindex_hashsz
);
109 while (pack_revindex
[i
].p
) {
110 if (pack_revindex
[i
].p
== p
)
112 if (++i
== pack_revindex_hashsz
)
118 static void prepare_pack_ix(void)
121 struct packed_git
*p
;
122 for (num
= 0, p
= packed_git
; p
; p
= p
->next
)
126 pack_revindex_hashsz
= num
* 11;
127 pack_revindex
= xcalloc(sizeof(*pack_revindex
), pack_revindex_hashsz
);
128 for (p
= packed_git
; p
; p
= p
->next
) {
129 num
= pack_revindex_ix(p
);
131 pack_revindex
[num
].p
= p
;
133 /* revindex elements are lazily initialized */
136 static int cmp_offset(const void *a_
, const void *b_
)
138 unsigned long a
= *(unsigned long *) a_
;
139 unsigned long b
= *(unsigned long *) b_
;
149 * Ordered list of offsets of objects in the pack.
151 static void prepare_pack_revindex(struct pack_revindex
*rix
)
153 struct packed_git
*p
= rix
->p
;
154 int num_ent
= num_packed_objects(p
);
156 void *index
= p
->index_base
+ 256;
158 rix
->revindex
= xmalloc(sizeof(unsigned long) * (num_ent
+ 1));
159 for (i
= 0; i
< num_ent
; i
++) {
160 uint32_t hl
= *((uint32_t *)(index
+ 24 * i
));
161 rix
->revindex
[i
] = ntohl(hl
);
163 /* This knows the pack format -- the 20-byte trailer
164 * follows immediately after the last object data.
166 rix
->revindex
[num_ent
] = p
->pack_size
- 20;
167 qsort(rix
->revindex
, num_ent
, sizeof(unsigned long), cmp_offset
);
170 static unsigned long find_packed_object_size(struct packed_git
*p
,
175 struct pack_revindex
*rix
;
176 unsigned long *revindex
;
177 num
= pack_revindex_ix(p
);
179 die("internal error: pack revindex uninitialized");
180 rix
= &pack_revindex
[num
];
182 prepare_pack_revindex(rix
);
183 revindex
= rix
->revindex
;
185 hi
= num_packed_objects(p
) + 1;
187 int mi
= (lo
+ hi
) / 2;
188 if (revindex
[mi
] == ofs
) {
189 return revindex
[mi
+1] - ofs
;
191 else if (ofs
< revindex
[mi
])
196 die("internal error: pack revindex corrupt");
199 static void *delta_against(void *buf
, unsigned long size
, struct object_entry
*entry
)
201 unsigned long othersize
, delta_size
;
203 void *otherbuf
= read_sha1_file(entry
->delta
->sha1
, type
, &othersize
);
207 die("unable to read %s", sha1_to_hex(entry
->delta
->sha1
));
208 delta_buf
= diff_delta(otherbuf
, othersize
,
209 buf
, size
, &delta_size
, 0);
210 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
211 die("delta size changed");
218 * The per-object header is a pretty dense thing, which is
219 * - first byte: low four bits are "size", then three bits of "type",
220 * and the high bit is "size continues".
221 * - each byte afterwards: low seven bits are size continuation,
222 * with the high bit being "size continues"
224 static int encode_header(enum object_type type
, unsigned long size
, unsigned char *hdr
)
229 if (type
< OBJ_COMMIT
|| type
> OBJ_DELTA
)
230 die("bad type %d", type
);
232 c
= (type
<< 4) | (size
& 15);
244 static unsigned long write_object(struct sha1file
*f
,
245 struct object_entry
*entry
)
250 unsigned char header
[10];
251 unsigned hdrlen
, datalen
;
252 enum object_type obj_type
;
255 if (entry
->preferred_base
)
258 obj_type
= entry
->type
;
259 if (! entry
->in_pack
)
260 to_reuse
= 0; /* can't reuse what we don't have */
261 else if (obj_type
== OBJ_DELTA
)
262 to_reuse
= 1; /* check_object() decided it for us */
263 else if (obj_type
!= entry
->in_pack_type
)
264 to_reuse
= 0; /* pack has delta which is unusable */
265 else if (entry
->delta
)
266 to_reuse
= 0; /* we want to pack afresh */
268 to_reuse
= 1; /* we have it in-pack undeltified,
269 * and we do not need to deltify it.
273 buf
= read_sha1_file(entry
->sha1
, type
, &size
);
275 die("unable to read %s", sha1_to_hex(entry
->sha1
));
276 if (size
!= entry
->size
)
277 die("object %s size inconsistency (%lu vs %lu)",
278 sha1_to_hex(entry
->sha1
), size
, entry
->size
);
280 buf
= delta_against(buf
, size
, entry
);
281 size
= entry
->delta_size
;
282 obj_type
= OBJ_DELTA
;
285 * The object header is a byte of 'type' followed by zero or
286 * more bytes of length. For deltas, the 20 bytes of delta
289 hdrlen
= encode_header(obj_type
, size
, header
);
290 sha1write(f
, header
, hdrlen
);
293 sha1write(f
, entry
->delta
, 20);
296 datalen
= sha1write_compressed(f
, buf
, size
);
300 struct packed_git
*p
= entry
->in_pack
;
303 datalen
= find_packed_object_size(p
, entry
->in_pack_offset
);
304 buf
= p
->pack_base
+ entry
->in_pack_offset
;
305 sha1write(f
, buf
, datalen
);
307 hdrlen
= 0; /* not really */
308 if (obj_type
== OBJ_DELTA
)
312 if (obj_type
== OBJ_DELTA
)
315 return hdrlen
+ datalen
;
318 static unsigned long write_one(struct sha1file
*f
,
319 struct object_entry
*e
,
320 unsigned long offset
)
323 /* offset starts from header size and cannot be zero
324 * if it is written already.
328 offset
+= write_object(f
, e
);
329 /* if we are deltified, write out its base object. */
331 offset
= write_one(f
, e
->delta
, offset
);
335 static void write_pack_file(void)
339 unsigned long offset
;
340 struct pack_header hdr
;
341 unsigned last_percent
= 999;
345 f
= sha1fd(1, "<stdout>");
347 f
= sha1create("%s-%s.%s", base_name
,
348 sha1_to_hex(object_list_sha1
), "pack");
349 do_progress
= progress
;
352 fprintf(stderr
, "Writing %d objects.\n", nr_result
);
354 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
355 hdr
.hdr_version
= htonl(PACK_VERSION
);
356 hdr
.hdr_entries
= htonl(nr_result
);
357 sha1write(f
, &hdr
, sizeof(hdr
));
358 offset
= sizeof(hdr
);
361 for (i
= 0; i
< nr_objects
; i
++) {
362 offset
= write_one(f
, objects
+ i
, offset
);
364 unsigned percent
= written
* 100 / nr_result
;
365 if (progress_update
|| percent
!= last_percent
) {
366 fprintf(stderr
, "%4u%% (%u/%u) done\r",
367 percent
, written
, nr_result
);
369 last_percent
= percent
;
376 sha1close(f
, pack_file_sha1
, 1);
379 static void write_index_file(void)
382 struct sha1file
*f
= sha1create("%s-%s.%s", base_name
,
383 sha1_to_hex(object_list_sha1
), "idx");
384 struct object_entry
**list
= sorted_by_sha
;
385 struct object_entry
**last
= list
+ nr_result
;
386 unsigned int array
[256];
389 * Write the first-level table (the list is sorted,
390 * but we use a 256-entry lookup to be able to avoid
391 * having to do eight extra binary search iterations).
393 for (i
= 0; i
< 256; i
++) {
394 struct object_entry
**next
= list
;
395 while (next
< last
) {
396 struct object_entry
*entry
= *next
;
397 if (entry
->sha1
[0] != i
)
401 array
[i
] = htonl(next
- sorted_by_sha
);
404 sha1write(f
, array
, 256 * sizeof(int));
407 * Write the actual SHA1 entries..
409 list
= sorted_by_sha
;
410 for (i
= 0; i
< nr_result
; i
++) {
411 struct object_entry
*entry
= *list
++;
412 unsigned int offset
= htonl(entry
->offset
);
413 sha1write(f
, &offset
, 4);
414 sha1write(f
, entry
->sha1
, 20);
416 sha1write(f
, pack_file_sha1
, 20);
417 sha1close(f
, NULL
, 1);
420 static int locate_object_entry_hash(const unsigned char *sha1
)
424 memcpy(&ui
, sha1
, sizeof(unsigned int));
425 i
= ui
% object_ix_hashsz
;
426 while (0 < object_ix
[i
]) {
427 if (!memcmp(sha1
, objects
[object_ix
[i
]-1].sha1
, 20))
429 if (++i
== object_ix_hashsz
)
435 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
439 if (!object_ix_hashsz
)
442 i
= locate_object_entry_hash(sha1
);
444 return &objects
[object_ix
[i
]-1];
448 static void rehash_objects(void)
451 struct object_entry
*oe
;
453 object_ix_hashsz
= nr_objects
* 3;
454 if (object_ix_hashsz
< 1024)
455 object_ix_hashsz
= 1024;
456 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
457 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
458 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
459 int ix
= locate_object_entry_hash(oe
->sha1
);
463 object_ix
[ix
] = i
+ 1;
468 struct name_path
*up
;
475 static unsigned name_hash(struct name_path
*path
, const char *name
)
477 struct name_path
*p
= path
;
478 const char *n
= name
+ strlen(name
);
479 unsigned hash
= 0, name_hash
= 0, name_done
= 0;
481 if (n
!= name
&& n
[-1] == '\n')
483 while (name
<= --n
) {
484 unsigned char c
= *n
;
485 if (c
== '/' && !name_done
) {
490 hash
= hash
* 11 + c
;
496 for (p
= path
; p
; p
= p
->up
) {
497 hash
= hash
* 11 + '/';
498 n
= p
->elem
+ p
->len
;
499 while (p
->elem
<= --n
) {
500 unsigned char c
= *n
;
501 hash
= hash
* 11 + c
;
505 * Make sure "Makefile" and "t/Makefile" are hashed separately
508 hash
= (name_hash
<<DIRBITS
) | (hash
& ((1U<<DIRBITS
)-1));
512 static int add_object_entry(const unsigned char *sha1
, unsigned hash
, int exclude
)
514 unsigned int idx
= nr_objects
;
515 struct object_entry
*entry
;
516 struct packed_git
*p
;
517 unsigned int found_offset
= 0;
518 struct packed_git
*found_pack
= NULL
;
522 for (p
= packed_git
; p
; p
= p
->next
) {
524 if (find_pack_entry_one(sha1
, &e
, p
)) {
527 if (local
&& !p
->pack_local
)
530 found_offset
= e
.offset
;
536 if ((entry
= locate_object_entry(sha1
)) != NULL
)
539 if (idx
>= nr_alloc
) {
540 unsigned int needed
= (idx
+ 1024) * 3 / 2;
541 objects
= xrealloc(objects
, needed
* sizeof(*entry
));
544 entry
= objects
+ idx
;
545 nr_objects
= idx
+ 1;
546 memset(entry
, 0, sizeof(*entry
));
547 memcpy(entry
->sha1
, sha1
, 20);
550 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
553 ix
= locate_object_entry_hash(entry
->sha1
);
555 die("internal error in object hashing.");
556 object_ix
[-1 - ix
] = idx
+ 1;
561 if (progress_update
) {
562 fprintf(stderr
, "Counting objects...%d\r", nr_objects
);
566 entry
->preferred_base
= 1;
569 entry
->in_pack
= found_pack
;
570 entry
->in_pack_offset
= found_offset
;
576 struct pbase_tree_cache
{
577 unsigned char sha1
[20];
581 unsigned long tree_size
;
584 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
585 static int pbase_tree_cache_ix(const unsigned char *sha1
)
587 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
589 static int pbase_tree_cache_ix_incr(int ix
)
591 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
594 static struct pbase_tree
{
595 struct pbase_tree
*next
;
596 /* This is a phony "cache" entry; we are not
597 * going to evict it nor find it through _get()
598 * mechanism -- this is for the toplevel node that
599 * would almost always change with any commit.
601 struct pbase_tree_cache pcache
;
604 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
606 struct pbase_tree_cache
*ent
, *nent
;
611 int my_ix
= pbase_tree_cache_ix(sha1
);
612 int available_ix
= -1;
614 /* pbase-tree-cache acts as a limited hashtable.
615 * your object will be found at your index or within a few
616 * slots after that slot if it is cached.
618 for (neigh
= 0; neigh
< 8; neigh
++) {
619 ent
= pbase_tree_cache
[my_ix
];
620 if (ent
&& !memcmp(ent
->sha1
, sha1
, 20)) {
624 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
625 ((0 <= available_ix
) &&
626 (!ent
&& pbase_tree_cache
[available_ix
])))
627 available_ix
= my_ix
;
630 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
633 /* Did not find one. Either we got a bogus request or
634 * we need to read and perhaps cache.
636 data
= read_sha1_file(sha1
, type
, &size
);
639 if (strcmp(type
, tree_type
)) {
644 /* We need to either cache or return a throwaway copy */
646 if (available_ix
< 0)
649 ent
= pbase_tree_cache
[available_ix
];
650 my_ix
= available_ix
;
654 nent
= xmalloc(sizeof(*nent
));
655 nent
->temporary
= (available_ix
< 0);
658 /* evict and reuse */
659 free(ent
->tree_data
);
662 memcpy(nent
->sha1
, sha1
, 20);
663 nent
->tree_data
= data
;
664 nent
->tree_size
= size
;
666 if (!nent
->temporary
)
667 pbase_tree_cache
[my_ix
] = nent
;
671 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
673 if (!cache
->temporary
) {
677 free(cache
->tree_data
);
681 static int name_cmp_len(const char *name
)
684 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
689 static void add_pbase_object(struct tree_desc
*tree
,
690 struct name_path
*up
,
695 const unsigned char *sha1
;
696 const char *entry_name
;
702 sha1
= tree_entry_extract(tree
, &entry_name
, &mode
);
703 update_tree_entry(tree
);
704 entry_len
= strlen(entry_name
);
705 if (entry_len
!= cmplen
||
706 memcmp(entry_name
, name
, cmplen
) ||
707 !has_sha1_file(sha1
) ||
708 sha1_object_info(sha1
, type
, &size
))
710 if (name
[cmplen
] != '/') {
711 unsigned hash
= name_hash(up
, name
);
712 add_object_entry(sha1
, hash
, 1);
715 if (!strcmp(type
, tree_type
)) {
716 struct tree_desc sub
;
718 struct pbase_tree_cache
*tree
;
719 const char *down
= name
+cmplen
+1;
720 int downlen
= name_cmp_len(down
);
722 tree
= pbase_tree_get(sha1
);
725 sub
.buf
= tree
->tree_data
;
726 sub
.size
= tree
->tree_size
;
729 me
.elem
= entry_name
;
731 add_pbase_object(&sub
, &me
, down
, downlen
);
732 pbase_tree_put(tree
);
737 static unsigned *done_pbase_paths
;
738 static int done_pbase_paths_num
;
739 static int done_pbase_paths_alloc
;
740 static int done_pbase_path_pos(unsigned hash
)
743 int hi
= done_pbase_paths_num
;
745 int mi
= (hi
+ lo
) / 2;
746 if (done_pbase_paths
[mi
] == hash
)
748 if (done_pbase_paths
[mi
] < hash
)
756 static int check_pbase_path(unsigned hash
)
758 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
762 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
763 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
764 done_pbase_paths
= xrealloc(done_pbase_paths
,
765 done_pbase_paths_alloc
*
768 done_pbase_paths_num
++;
769 if (pos
< done_pbase_paths_num
)
770 memmove(done_pbase_paths
+ pos
+ 1,
771 done_pbase_paths
+ pos
,
772 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
773 done_pbase_paths
[pos
] = hash
;
777 static void add_preferred_base_object(char *name
, unsigned hash
)
779 struct pbase_tree
*it
;
780 int cmplen
= name_cmp_len(name
);
782 if (check_pbase_path(hash
))
785 for (it
= pbase_tree
; it
; it
= it
->next
) {
787 hash
= name_hash(NULL
, "");
788 add_object_entry(it
->pcache
.sha1
, hash
, 1);
791 struct tree_desc tree
;
792 tree
.buf
= it
->pcache
.tree_data
;
793 tree
.size
= it
->pcache
.tree_size
;
794 add_pbase_object(&tree
, NULL
, name
, cmplen
);
799 static void add_preferred_base(unsigned char *sha1
)
801 struct pbase_tree
*it
;
804 unsigned char tree_sha1
[20];
806 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
810 for (it
= pbase_tree
; it
; it
= it
->next
) {
811 if (!memcmp(it
->pcache
.sha1
, tree_sha1
, 20)) {
817 it
= xcalloc(1, sizeof(*it
));
818 it
->next
= pbase_tree
;
821 memcpy(it
->pcache
.sha1
, tree_sha1
, 20);
822 it
->pcache
.tree_data
= data
;
823 it
->pcache
.tree_size
= size
;
826 static void check_object(struct object_entry
*entry
)
830 if (entry
->in_pack
&& !entry
->preferred_base
) {
831 unsigned char base
[20];
833 struct object_entry
*base_entry
;
835 /* We want in_pack_type even if we do not reuse delta.
836 * There is no point not reusing non-delta representations.
838 check_reuse_pack_delta(entry
->in_pack
,
839 entry
->in_pack_offset
,
841 &entry
->in_pack_type
);
843 /* Check if it is delta, and the base is also an object
844 * we are going to pack. If so we will reuse the existing
847 if (!no_reuse_delta
&&
848 entry
->in_pack_type
== OBJ_DELTA
&&
849 (base_entry
= locate_object_entry(base
)) &&
850 (!base_entry
->preferred_base
)) {
852 /* Depth value does not matter - find_deltas()
853 * will never consider reused delta as the
854 * base object to deltify other objects
855 * against, in order to avoid circular deltas.
858 /* uncompressed size of the delta data */
859 entry
->size
= entry
->delta_size
= size
;
860 entry
->delta
= base_entry
;
861 entry
->type
= OBJ_DELTA
;
863 entry
->delta_sibling
= base_entry
->delta_child
;
864 base_entry
->delta_child
= entry
;
868 /* Otherwise we would do the usual */
871 if (sha1_object_info(entry
->sha1
, type
, &entry
->size
))
872 die("unable to get type of object %s",
873 sha1_to_hex(entry
->sha1
));
875 if (!strcmp(type
, commit_type
)) {
876 entry
->type
= OBJ_COMMIT
;
877 } else if (!strcmp(type
, tree_type
)) {
878 entry
->type
= OBJ_TREE
;
879 } else if (!strcmp(type
, blob_type
)) {
880 entry
->type
= OBJ_BLOB
;
881 } else if (!strcmp(type
, tag_type
)) {
882 entry
->type
= OBJ_TAG
;
884 die("unable to pack object %s of type %s",
885 sha1_to_hex(entry
->sha1
), type
);
888 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
890 struct object_entry
*child
= me
->delta_child
;
893 unsigned int c
= check_delta_limit(child
, n
+ 1);
896 child
= child
->delta_sibling
;
901 static void get_object_details(void)
904 struct object_entry
*entry
;
907 for (i
= 0, entry
= objects
; i
< nr_objects
; i
++, entry
++)
910 if (nr_objects
== nr_result
) {
912 * Depth of objects that depend on the entry -- this
913 * is subtracted from depth-max to break too deep
914 * delta chain because of delta data reusing.
915 * However, we loosen this restriction when we know we
916 * are creating a thin pack -- it will have to be
917 * expanded on the other end anyway, so do not
918 * artificially cut the delta chain and let it go as
921 for (i
= 0, entry
= objects
; i
< nr_objects
; i
++, entry
++)
922 if (!entry
->delta
&& entry
->delta_child
)
924 check_delta_limit(entry
, 1);
928 typedef int (*entry_sort_t
)(const struct object_entry
*, const struct object_entry
*);
930 static entry_sort_t current_sort
;
932 static int sort_comparator(const void *_a
, const void *_b
)
934 struct object_entry
*a
= *(struct object_entry
**)_a
;
935 struct object_entry
*b
= *(struct object_entry
**)_b
;
936 return current_sort(a
,b
);
939 static struct object_entry
**create_sorted_list(entry_sort_t sort
)
941 struct object_entry
**list
= xmalloc(nr_objects
* sizeof(struct object_entry
*));
944 for (i
= 0; i
< nr_objects
; i
++)
945 list
[i
] = objects
+ i
;
947 qsort(list
, nr_objects
, sizeof(struct object_entry
*), sort_comparator
);
951 static int sha1_sort(const struct object_entry
*a
, const struct object_entry
*b
)
953 return memcmp(a
->sha1
, b
->sha1
, 20);
956 static struct object_entry
**create_final_object_list(void)
958 struct object_entry
**list
;
961 for (i
= nr_result
= 0; i
< nr_objects
; i
++)
962 if (!objects
[i
].preferred_base
)
964 list
= xmalloc(nr_result
* sizeof(struct object_entry
*));
965 for (i
= j
= 0; i
< nr_objects
; i
++) {
966 if (!objects
[i
].preferred_base
)
967 list
[j
++] = objects
+ i
;
969 current_sort
= sha1_sort
;
970 qsort(list
, nr_result
, sizeof(struct object_entry
*), sort_comparator
);
974 static int type_size_sort(const struct object_entry
*a
, const struct object_entry
*b
)
976 if (a
->type
< b
->type
)
978 if (a
->type
> b
->type
)
980 if (a
->hash
< b
->hash
)
982 if (a
->hash
> b
->hash
)
984 if (a
->preferred_base
< b
->preferred_base
)
986 if (a
->preferred_base
> b
->preferred_base
)
988 if (a
->size
< b
->size
)
990 if (a
->size
> b
->size
)
992 return a
< b
? -1 : (a
> b
);
996 struct object_entry
*entry
;
998 struct delta_index
*index
;
1002 * We search for deltas _backwards_ in a list sorted by type and
1003 * by size, so that we see progressively smaller and smaller files.
1004 * That's because we prefer deltas to be from the bigger file
1005 * to the smaller - deletes are potentially cheaper, but perhaps
1006 * more importantly, the bigger file is likely the more recent
1009 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1010 struct delta_index
*src_index
, unsigned max_depth
)
1012 struct object_entry
*trg_entry
= trg
->entry
;
1013 struct object_entry
*src_entry
= src
->entry
;
1014 unsigned long size
, src_size
, delta_size
, sizediff
, max_size
;
1017 /* Don't bother doing diffs between different types */
1018 if (trg_entry
->type
!= src_entry
->type
)
1021 /* We do not compute delta to *create* objects we are not
1024 if (trg_entry
->preferred_base
)
1028 * If the current object is at pack edge, take the depth the
1029 * objects that depend on the current object into account --
1030 * otherwise they would become too deep.
1032 if (trg_entry
->delta_child
) {
1033 if (max_depth
<= trg_entry
->delta_limit
)
1035 max_depth
-= trg_entry
->delta_limit
;
1037 if (src_entry
->depth
>= max_depth
)
1040 /* Now some size filtering euristics. */
1041 size
= trg_entry
->size
;
1042 max_size
= (size
/2 - 20) / (src_entry
->depth
+ 1);
1043 if (trg_entry
->delta
&& trg_entry
->delta_size
<= max_size
)
1044 max_size
= trg_entry
->delta_size
-1;
1045 src_size
= src_entry
->size
;
1046 sizediff
= src_size
< size
? size
- src_size
: 0;
1047 if (sizediff
>= max_size
)
1050 delta_buf
= create_delta(src_index
, trg
->data
, size
, &delta_size
, max_size
);
1054 trg_entry
->delta
= src_entry
;
1055 trg_entry
->delta_size
= delta_size
;
1056 trg_entry
->depth
= src_entry
->depth
+ 1;
1061 static void progress_interval(int signum
)
1063 progress_update
= 1;
1066 static void find_deltas(struct object_entry
**list
, int window
, int depth
)
1069 unsigned int array_size
= window
* sizeof(struct unpacked
);
1070 struct unpacked
*array
= xmalloc(array_size
);
1071 unsigned processed
= 0;
1072 unsigned last_percent
= 999;
1074 memset(array
, 0, array_size
);
1078 fprintf(stderr
, "Deltifying %d objects.\n", nr_result
);
1081 struct object_entry
*entry
= list
[i
];
1082 struct unpacked
*n
= array
+ idx
;
1087 if (!entry
->preferred_base
)
1091 unsigned percent
= processed
* 100 / nr_result
;
1092 if (percent
!= last_percent
|| progress_update
) {
1093 fprintf(stderr
, "%4u%% (%u/%u) done\r",
1094 percent
, processed
, nr_result
);
1095 progress_update
= 0;
1096 last_percent
= percent
;
1101 /* This happens if we decided to reuse existing
1102 * delta from a pack. "!no_reuse_delta &&" is implied.
1106 if (entry
->size
< 50)
1108 free_delta_index(n
->index
);
1112 n
->data
= read_sha1_file(entry
->sha1
, type
, &size
);
1113 if (size
!= entry
->size
)
1114 die("object %s inconsistent object length (%lu vs %lu)",
1115 sha1_to_hex(entry
->sha1
), size
, entry
->size
);
1119 unsigned int other_idx
= idx
+ j
;
1121 if (other_idx
>= window
)
1122 other_idx
-= window
;
1123 m
= array
+ other_idx
;
1126 if (try_delta(n
, m
, m
->index
, depth
) < 0)
1129 /* if we made n a delta, and if n is already at max
1130 * depth, leaving it in the window is pointless. we
1131 * should evict it first.
1133 if (entry
->delta
&& depth
<= entry
->depth
)
1136 n
->index
= create_delta_index(n
->data
, size
);
1138 die("out of memory");
1146 fputc('\n', stderr
);
1148 for (i
= 0; i
< window
; ++i
) {
1149 free_delta_index(array
[i
].index
);
1150 free(array
[i
].data
);
1155 static void prepare_pack(int window
, int depth
)
1157 get_object_details();
1158 sorted_by_type
= create_sorted_list(type_size_sort
);
1159 if (window
&& depth
)
1160 find_deltas(sorted_by_type
, window
+1, depth
);
1163 static int reuse_cached_pack(unsigned char *sha1
, int pack_to_stdout
)
1165 static const char cache
[] = "pack-cache/pack-%s.%s";
1166 char *cached_pack
, *cached_idx
;
1167 int ifd
, ofd
, ifd_ix
= -1;
1169 cached_pack
= git_path(cache
, sha1_to_hex(sha1
), "pack");
1170 ifd
= open(cached_pack
, O_RDONLY
);
1174 if (!pack_to_stdout
) {
1175 cached_idx
= git_path(cache
, sha1_to_hex(sha1
), "idx");
1176 ifd_ix
= open(cached_idx
, O_RDONLY
);
1184 fprintf(stderr
, "Reusing %d objects pack %s\n", nr_objects
,
1187 if (pack_to_stdout
) {
1188 if (copy_fd(ifd
, 1))
1193 char name
[PATH_MAX
];
1194 snprintf(name
, sizeof(name
),
1195 "%s-%s.%s", base_name
, sha1_to_hex(sha1
), "pack");
1196 ofd
= open(name
, O_CREAT
| O_EXCL
| O_WRONLY
, 0666);
1198 die("unable to open %s (%s)", name
, strerror(errno
));
1199 if (copy_fd(ifd
, ofd
))
1203 snprintf(name
, sizeof(name
),
1204 "%s-%s.%s", base_name
, sha1_to_hex(sha1
), "idx");
1205 ofd
= open(name
, O_CREAT
| O_EXCL
| O_WRONLY
, 0666);
1207 die("unable to open %s (%s)", name
, strerror(errno
));
1208 if (copy_fd(ifd_ix
, ofd
))
1211 puts(sha1_to_hex(sha1
));
1217 static void setup_progress_signal(void)
1219 struct sigaction sa
;
1222 memset(&sa
, 0, sizeof(sa
));
1223 sa
.sa_handler
= progress_interval
;
1224 sigemptyset(&sa
.sa_mask
);
1225 sa
.sa_flags
= SA_RESTART
;
1226 sigaction(SIGALRM
, &sa
, NULL
);
1228 v
.it_interval
.tv_sec
= 1;
1229 v
.it_interval
.tv_usec
= 0;
1230 v
.it_value
= v
.it_interval
;
1231 setitimer(ITIMER_REAL
, &v
, NULL
);
1234 int main(int argc
, char **argv
)
1237 char line
[40 + 1 + PATH_MAX
+ 2];
1238 int window
= 10, depth
= 10, pack_to_stdout
= 0;
1239 struct object_entry
**list
;
1240 int num_preferred_base
= 0;
1243 setup_git_directory();
1245 progress
= isatty(2);
1246 for (i
= 1; i
< argc
; i
++) {
1247 const char *arg
= argv
[i
];
1250 if (!strcmp("--non-empty", arg
)) {
1254 if (!strcmp("--local", arg
)) {
1258 if (!strcmp("--incremental", arg
)) {
1262 if (!strncmp("--window=", arg
, 9)) {
1264 window
= strtoul(arg
+9, &end
, 0);
1265 if (!arg
[9] || *end
)
1269 if (!strncmp("--depth=", arg
, 8)) {
1271 depth
= strtoul(arg
+8, &end
, 0);
1272 if (!arg
[8] || *end
)
1276 if (!strcmp("--progress", arg
)) {
1280 if (!strcmp("-q", arg
)) {
1284 if (!strcmp("--no-reuse-delta", arg
)) {
1288 if (!strcmp("--stdout", arg
)) {
1299 if (pack_to_stdout
!= !base_name
)
1302 prepare_packed_git();
1305 fprintf(stderr
, "Generating pack...\n");
1306 setup_progress_signal();
1310 unsigned char sha1
[20];
1313 if (!fgets(line
, sizeof(line
), stdin
)) {
1317 die("fgets returned NULL, not EOF, not error!");
1319 die("fgets: %s", strerror(errno
));
1324 if (line
[0] == '-') {
1325 if (get_sha1_hex(line
+1, sha1
))
1326 die("expected edge sha1, got garbage:\n %s",
1328 if (num_preferred_base
++ < window
)
1329 add_preferred_base(sha1
);
1332 if (get_sha1_hex(line
, sha1
))
1333 die("expected sha1, got garbage:\n %s", line
);
1334 hash
= name_hash(NULL
, line
+41);
1335 add_preferred_base_object(line
+41, hash
);
1336 add_object_entry(sha1
, hash
, 0);
1339 fprintf(stderr
, "Done counting %d objects.\n", nr_objects
);
1340 sorted_by_sha
= create_final_object_list();
1341 if (non_empty
&& !nr_result
)
1345 list
= sorted_by_sha
;
1346 for (i
= 0; i
< nr_result
; i
++) {
1347 struct object_entry
*entry
= *list
++;
1348 SHA1_Update(&ctx
, entry
->sha1
, 20);
1350 SHA1_Final(object_list_sha1
, &ctx
);
1351 if (progress
&& (nr_objects
!= nr_result
))
1352 fprintf(stderr
, "Result has %d objects.\n", nr_result
);
1354 if (reuse_cached_pack(object_list_sha1
, pack_to_stdout
))
1358 prepare_pack(window
, depth
);
1359 if (progress
&& pack_to_stdout
) {
1360 /* the other end usually displays progress itself */
1361 struct itimerval v
= {{0,},};
1362 setitimer(ITIMER_REAL
, &v
, NULL
);
1363 signal(SIGALRM
, SIG_IGN
);
1364 progress_update
= 0;
1367 if (!pack_to_stdout
) {
1369 puts(sha1_to_hex(object_list_sha1
));
1373 fprintf(stderr
, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1374 nr_result
, written
, written_delta
, reused
, reused_delta
);