10 #include "csum-file.h"
11 #include "tree-walk.h"
14 #include "list-objects.h"
18 static const char pack_usage
[] = "\
19 git-pack-objects [{ -q | --progress | --all-progress }] \n\
20 [--local] [--incremental] [--window=N] [--depth=N] \n\
21 [--no-reuse-delta] [--delta-base-offset] [--non-empty] \n\
22 [--revs [--unpacked | --all]*] [--stdout | base-name] \n\
23 [<ref-list | <object-list]";
26 unsigned char sha1
[20];
27 unsigned long size
; /* uncompressed size */
28 unsigned long offset
; /* offset into the final pack file;
29 * nonzero if already written.
31 unsigned int depth
; /* delta depth */
32 unsigned int delta_limit
; /* base adjustment for in-pack delta */
33 unsigned int hash
; /* name hint hash */
34 enum object_type type
;
35 enum object_type in_pack_type
; /* could be delta */
36 unsigned long delta_size
; /* delta data size (uncompressed) */
37 #define in_pack_header_size delta_size /* only when reusing pack data */
38 struct object_entry
*delta
; /* delta base object */
39 struct packed_git
*in_pack
; /* already in pack */
40 unsigned int in_pack_offset
;
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 int preferred_base
; /* we do not pack this, but is encouraged to
46 * be used as the base objectto delta huge
52 * Objects we are going to pack are collected in objects array (dynamically
53 * expanded). nr_objects & nr_alloc controls this array. They are stored
54 * in the order we see -- typically rev-list --objects order that gives us
55 * nice "minimum seek" order.
57 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
58 * elements in the objects array. The former is used to build the pack
59 * index (lists object names in the ascending order to help offset lookup),
60 * and the latter is used to group similar things together by try_delta()
64 static unsigned char object_list_sha1
[20];
66 static int no_reuse_delta
;
68 static int incremental
;
69 static int allow_ofs_delta
;
71 static struct object_entry
**sorted_by_sha
, **sorted_by_type
;
72 static struct object_entry
*objects
;
73 static int nr_objects
, nr_alloc
, nr_result
;
74 static const char *base_name
;
75 static unsigned char pack_file_sha1
[20];
76 static int progress
= 1;
77 static volatile sig_atomic_t progress_update
;
78 static int window
= 10;
79 static int pack_to_stdout
;
80 static int num_preferred_base
;
83 * The object names in objects array are hashed with this hashtable,
84 * to help looking up the entry by object name. Binary search from
85 * sorted_by_sha is also possible but this was easier to code and faster.
86 * This hashtable is built after all the objects are seen.
88 static int *object_ix
;
89 static int object_ix_hashsz
;
92 * Pack index for existing packs give us easy access to the offsets into
93 * corresponding pack file where each object's data starts, but the entries
94 * do not store the size of the compressed representation (uncompressed
95 * size is easily available by examining the pack entry header). It is
96 * also rather expensive to find the sha1 for an object given its offset.
98 * We build a hashtable of existing packs (pack_revindex), and keep reverse
99 * index here -- pack index file is sorted by object name mapping to offset;
100 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
101 * ordered by offset, so if you know the offset of an object, next offset
102 * is where its packed representation ends and the index_nr can be used to
103 * get the object sha1 from the main index.
105 struct revindex_entry
{
109 struct pack_revindex
{
110 struct packed_git
*p
;
111 struct revindex_entry
*revindex
;
113 static struct pack_revindex
*pack_revindex
;
114 static int pack_revindex_hashsz
;
120 static int written_delta
;
122 static int reused_delta
;
124 static int pack_revindex_ix(struct packed_git
*p
)
126 unsigned long ui
= (unsigned long)p
;
129 ui
= ui
^ (ui
>> 16); /* defeat structure alignment */
130 i
= (int)(ui
% pack_revindex_hashsz
);
131 while (pack_revindex
[i
].p
) {
132 if (pack_revindex
[i
].p
== p
)
134 if (++i
== pack_revindex_hashsz
)
140 static void prepare_pack_ix(void)
143 struct packed_git
*p
;
144 for (num
= 0, p
= packed_git
; p
; p
= p
->next
)
148 pack_revindex_hashsz
= num
* 11;
149 pack_revindex
= xcalloc(sizeof(*pack_revindex
), pack_revindex_hashsz
);
150 for (p
= packed_git
; p
; p
= p
->next
) {
151 num
= pack_revindex_ix(p
);
153 pack_revindex
[num
].p
= p
;
155 /* revindex elements are lazily initialized */
158 static int cmp_offset(const void *a_
, const void *b_
)
160 const struct revindex_entry
*a
= a_
;
161 const struct revindex_entry
*b
= b_
;
162 return (a
->offset
< b
->offset
) ? -1 : (a
->offset
> b
->offset
) ? 1 : 0;
166 * Ordered list of offsets of objects in the pack.
168 static void prepare_pack_revindex(struct pack_revindex
*rix
)
170 struct packed_git
*p
= rix
->p
;
171 int num_ent
= num_packed_objects(p
);
173 void *index
= p
->index_base
+ 256;
175 rix
->revindex
= xmalloc(sizeof(*rix
->revindex
) * (num_ent
+ 1));
176 for (i
= 0; i
< num_ent
; i
++) {
177 unsigned int hl
= *((unsigned int *)((char *) index
+ 24*i
));
178 rix
->revindex
[i
].offset
= ntohl(hl
);
179 rix
->revindex
[i
].nr
= i
;
181 /* This knows the pack format -- the 20-byte trailer
182 * follows immediately after the last object data.
184 rix
->revindex
[num_ent
].offset
= p
->pack_size
- 20;
185 rix
->revindex
[num_ent
].nr
= -1;
186 qsort(rix
->revindex
, num_ent
, sizeof(*rix
->revindex
), cmp_offset
);
189 static struct revindex_entry
* find_packed_object(struct packed_git
*p
,
194 struct pack_revindex
*rix
;
195 struct revindex_entry
*revindex
;
196 num
= pack_revindex_ix(p
);
198 die("internal error: pack revindex uninitialized");
199 rix
= &pack_revindex
[num
];
201 prepare_pack_revindex(rix
);
202 revindex
= rix
->revindex
;
204 hi
= num_packed_objects(p
) + 1;
206 int mi
= (lo
+ hi
) / 2;
207 if (revindex
[mi
].offset
== ofs
) {
208 return revindex
+ mi
;
210 else if (ofs
< revindex
[mi
].offset
)
215 die("internal error: pack revindex corrupt");
218 static unsigned long find_packed_object_size(struct packed_git
*p
,
221 struct revindex_entry
*entry
= find_packed_object(p
, ofs
);
222 return entry
[1].offset
- ofs
;
225 static unsigned char *find_packed_object_name(struct packed_git
*p
,
228 struct revindex_entry
*entry
= find_packed_object(p
, ofs
);
229 return (unsigned char *)(p
->index_base
+ 256) + 24 * entry
->nr
+ 4;
232 static void *delta_against(void *buf
, unsigned long size
, struct object_entry
*entry
)
234 unsigned long othersize
, delta_size
;
236 void *otherbuf
= read_sha1_file(entry
->delta
->sha1
, type
, &othersize
);
240 die("unable to read %s", sha1_to_hex(entry
->delta
->sha1
));
241 delta_buf
= diff_delta(otherbuf
, othersize
,
242 buf
, size
, &delta_size
, 0);
243 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
244 die("delta size changed");
251 * The per-object header is a pretty dense thing, which is
252 * - first byte: low four bits are "size", then three bits of "type",
253 * and the high bit is "size continues".
254 * - each byte afterwards: low seven bits are size continuation,
255 * with the high bit being "size continues"
257 static int encode_header(enum object_type type
, unsigned long size
, unsigned char *hdr
)
262 if (type
< OBJ_COMMIT
|| type
> OBJ_REF_DELTA
)
263 die("bad type %d", type
);
265 c
= (type
<< 4) | (size
& 15);
278 * we are going to reuse the existing object data as is. make
279 * sure it is not corrupt.
281 static int check_inflate(unsigned char *data
, unsigned long len
, unsigned long expect
)
284 unsigned char fakebuf
[4096];
287 memset(&stream
, 0, sizeof(stream
));
288 stream
.next_in
= data
;
289 stream
.avail_in
= len
;
290 stream
.next_out
= fakebuf
;
291 stream
.avail_out
= sizeof(fakebuf
);
292 inflateInit(&stream
);
295 st
= inflate(&stream
, Z_FINISH
);
296 if (st
== Z_STREAM_END
|| st
== Z_OK
) {
297 st
= (stream
.total_out
== expect
&&
298 stream
.total_in
== len
) ? 0 : -1;
301 if (st
!= Z_BUF_ERROR
) {
305 stream
.next_out
= fakebuf
;
306 stream
.avail_out
= sizeof(fakebuf
);
312 static int revalidate_loose_object(struct object_entry
*entry
,
314 unsigned long mapsize
)
316 /* we already know this is a loose object with new type header. */
317 enum object_type type
;
318 unsigned long size
, used
;
323 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
328 return check_inflate(map
, mapsize
, size
);
331 static unsigned long write_object(struct sha1file
*f
,
332 struct object_entry
*entry
)
337 unsigned char header
[10];
338 unsigned hdrlen
, datalen
;
339 enum object_type obj_type
;
342 obj_type
= entry
->type
;
343 if (! entry
->in_pack
)
344 to_reuse
= 0; /* can't reuse what we don't have */
345 else if (obj_type
== OBJ_REF_DELTA
|| obj_type
== OBJ_OFS_DELTA
)
346 to_reuse
= 1; /* check_object() decided it for us */
347 else if (obj_type
!= entry
->in_pack_type
)
348 to_reuse
= 0; /* pack has delta which is unusable */
349 else if (entry
->delta
)
350 to_reuse
= 0; /* we want to pack afresh */
352 to_reuse
= 1; /* we have it in-pack undeltified,
353 * and we do not need to deltify it.
356 if (!entry
->in_pack
&& !entry
->delta
) {
358 unsigned long mapsize
;
359 map
= map_sha1_file(entry
->sha1
, &mapsize
);
360 if (map
&& !legacy_loose_object(map
)) {
361 /* We can copy straight into the pack file */
362 if (revalidate_loose_object(entry
, map
, mapsize
))
363 die("corrupt loose object %s",
364 sha1_to_hex(entry
->sha1
));
365 sha1write(f
, map
, mapsize
);
366 munmap(map
, mapsize
);
372 munmap(map
, mapsize
);
376 buf
= read_sha1_file(entry
->sha1
, type
, &size
);
378 die("unable to read %s", sha1_to_hex(entry
->sha1
));
379 if (size
!= entry
->size
)
380 die("object %s size inconsistency (%lu vs %lu)",
381 sha1_to_hex(entry
->sha1
), size
, entry
->size
);
383 buf
= delta_against(buf
, size
, entry
);
384 size
= entry
->delta_size
;
385 obj_type
= (allow_ofs_delta
&& entry
->delta
->offset
) ?
386 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
389 * The object header is a byte of 'type' followed by zero or
390 * more bytes of length.
392 hdrlen
= encode_header(obj_type
, size
, header
);
393 sha1write(f
, header
, hdrlen
);
395 if (obj_type
== OBJ_OFS_DELTA
) {
397 * Deltas with relative base contain an additional
398 * encoding of the relative offset for the delta
399 * base from this object's position in the pack.
401 unsigned long ofs
= entry
->offset
- entry
->delta
->offset
;
402 unsigned pos
= sizeof(header
) - 1;
403 header
[pos
] = ofs
& 127;
405 header
[--pos
] = 128 | (--ofs
& 127);
406 sha1write(f
, header
+ pos
, sizeof(header
) - pos
);
407 hdrlen
+= sizeof(header
) - pos
;
408 } else if (obj_type
== OBJ_REF_DELTA
) {
410 * Deltas with a base reference contain
411 * an additional 20 bytes for the base sha1.
413 sha1write(f
, entry
->delta
->sha1
, 20);
416 datalen
= sha1write_compressed(f
, buf
, size
);
420 struct packed_git
*p
= entry
->in_pack
;
423 obj_type
= (allow_ofs_delta
&& entry
->delta
->offset
) ?
424 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
427 hdrlen
= encode_header(obj_type
, entry
->size
, header
);
428 sha1write(f
, header
, hdrlen
);
429 if (obj_type
== OBJ_OFS_DELTA
) {
430 unsigned long ofs
= entry
->offset
- entry
->delta
->offset
;
431 unsigned pos
= sizeof(header
) - 1;
432 header
[pos
] = ofs
& 127;
434 header
[--pos
] = 128 | (--ofs
& 127);
435 sha1write(f
, header
+ pos
, sizeof(header
) - pos
);
436 hdrlen
+= sizeof(header
) - pos
;
437 } else if (obj_type
== OBJ_REF_DELTA
) {
438 sha1write(f
, entry
->delta
->sha1
, 20);
443 buf
= (char *) p
->pack_base
444 + entry
->in_pack_offset
445 + entry
->in_pack_header_size
;
446 datalen
= find_packed_object_size(p
, entry
->in_pack_offset
)
447 - entry
->in_pack_header_size
;
448 if (!pack_to_stdout
&& check_inflate(buf
, datalen
, entry
->size
))
449 die("corrupt delta in pack %s", sha1_to_hex(entry
->sha1
));
450 sha1write(f
, buf
, datalen
);
457 return hdrlen
+ datalen
;
460 static unsigned long write_one(struct sha1file
*f
,
461 struct object_entry
*e
,
462 unsigned long offset
)
464 if (e
->offset
|| e
->preferred_base
)
465 /* offset starts from header size and cannot be zero
466 * if it is written already.
469 /* if we are deltified, write out its base object first. */
471 offset
= write_one(f
, e
->delta
, offset
);
473 return offset
+ write_object(f
, e
);
476 static void write_pack_file(void)
480 unsigned long offset
;
481 struct pack_header hdr
;
482 unsigned last_percent
= 999;
483 int do_progress
= progress
;
486 f
= sha1fd(1, "<stdout>");
490 f
= sha1create("%s-%s.%s", base_name
,
491 sha1_to_hex(object_list_sha1
), "pack");
493 fprintf(stderr
, "Writing %d objects.\n", nr_result
);
495 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
496 hdr
.hdr_version
= htonl(PACK_VERSION
);
497 hdr
.hdr_entries
= htonl(nr_result
);
498 sha1write(f
, &hdr
, sizeof(hdr
));
499 offset
= sizeof(hdr
);
502 for (i
= 0; i
< nr_objects
; i
++) {
503 offset
= write_one(f
, objects
+ i
, offset
);
505 unsigned percent
= written
* 100 / nr_result
;
506 if (progress_update
|| percent
!= last_percent
) {
507 fprintf(stderr
, "%4u%% (%u/%u) done\r",
508 percent
, written
, nr_result
);
510 last_percent
= percent
;
517 sha1close(f
, pack_file_sha1
, 1);
520 static void write_index_file(void)
523 struct sha1file
*f
= sha1create("%s-%s.%s", base_name
,
524 sha1_to_hex(object_list_sha1
), "idx");
525 struct object_entry
**list
= sorted_by_sha
;
526 struct object_entry
**last
= list
+ nr_result
;
527 unsigned int array
[256];
530 * Write the first-level table (the list is sorted,
531 * but we use a 256-entry lookup to be able to avoid
532 * having to do eight extra binary search iterations).
534 for (i
= 0; i
< 256; i
++) {
535 struct object_entry
**next
= list
;
536 while (next
< last
) {
537 struct object_entry
*entry
= *next
;
538 if (entry
->sha1
[0] != i
)
542 array
[i
] = htonl(next
- sorted_by_sha
);
545 sha1write(f
, array
, 256 * sizeof(int));
548 * Write the actual SHA1 entries..
550 list
= sorted_by_sha
;
551 for (i
= 0; i
< nr_result
; i
++) {
552 struct object_entry
*entry
= *list
++;
553 unsigned int offset
= htonl(entry
->offset
);
554 sha1write(f
, &offset
, 4);
555 sha1write(f
, entry
->sha1
, 20);
557 sha1write(f
, pack_file_sha1
, 20);
558 sha1close(f
, NULL
, 1);
561 static int locate_object_entry_hash(const unsigned char *sha1
)
565 memcpy(&ui
, sha1
, sizeof(unsigned int));
566 i
= ui
% object_ix_hashsz
;
567 while (0 < object_ix
[i
]) {
568 if (!hashcmp(sha1
, objects
[object_ix
[i
] - 1].sha1
))
570 if (++i
== object_ix_hashsz
)
576 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
580 if (!object_ix_hashsz
)
583 i
= locate_object_entry_hash(sha1
);
585 return &objects
[object_ix
[i
]-1];
589 static void rehash_objects(void)
592 struct object_entry
*oe
;
594 object_ix_hashsz
= nr_objects
* 3;
595 if (object_ix_hashsz
< 1024)
596 object_ix_hashsz
= 1024;
597 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
598 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
599 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
600 int ix
= locate_object_entry_hash(oe
->sha1
);
604 object_ix
[ix
] = i
+ 1;
608 static unsigned name_hash(const char *name
)
614 * This effectively just creates a sortable number from the
615 * last sixteen non-whitespace characters. Last characters
616 * count "most", so things that end in ".c" sort together.
618 while ((c
= *name
++) != 0) {
621 hash
= (hash
>> 2) + (c
<< 24);
626 static int add_object_entry(const unsigned char *sha1
, unsigned hash
, int exclude
)
628 unsigned int idx
= nr_objects
;
629 struct object_entry
*entry
;
630 struct packed_git
*p
;
631 unsigned int found_offset
= 0;
632 struct packed_git
*found_pack
= NULL
;
636 for (p
= packed_git
; p
; p
= p
->next
) {
637 unsigned long offset
= find_pack_entry_one(sha1
, p
);
641 if (local
&& !p
->pack_local
)
644 found_offset
= offset
;
650 if ((entry
= locate_object_entry(sha1
)) != NULL
)
653 if (idx
>= nr_alloc
) {
654 unsigned int needed
= (idx
+ 1024) * 3 / 2;
655 objects
= xrealloc(objects
, needed
* sizeof(*entry
));
658 entry
= objects
+ idx
;
659 nr_objects
= idx
+ 1;
660 memset(entry
, 0, sizeof(*entry
));
661 hashcpy(entry
->sha1
, sha1
);
664 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
667 ix
= locate_object_entry_hash(entry
->sha1
);
669 die("internal error in object hashing.");
670 object_ix
[-1 - ix
] = idx
+ 1;
675 if (progress_update
) {
676 fprintf(stderr
, "Counting objects...%d\r", nr_objects
);
680 entry
->preferred_base
= 1;
683 entry
->in_pack
= found_pack
;
684 entry
->in_pack_offset
= found_offset
;
690 struct pbase_tree_cache
{
691 unsigned char sha1
[20];
695 unsigned long tree_size
;
698 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
699 static int pbase_tree_cache_ix(const unsigned char *sha1
)
701 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
703 static int pbase_tree_cache_ix_incr(int ix
)
705 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
708 static struct pbase_tree
{
709 struct pbase_tree
*next
;
710 /* This is a phony "cache" entry; we are not
711 * going to evict it nor find it through _get()
712 * mechanism -- this is for the toplevel node that
713 * would almost always change with any commit.
715 struct pbase_tree_cache pcache
;
718 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
720 struct pbase_tree_cache
*ent
, *nent
;
725 int my_ix
= pbase_tree_cache_ix(sha1
);
726 int available_ix
= -1;
728 /* pbase-tree-cache acts as a limited hashtable.
729 * your object will be found at your index or within a few
730 * slots after that slot if it is cached.
732 for (neigh
= 0; neigh
< 8; neigh
++) {
733 ent
= pbase_tree_cache
[my_ix
];
734 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
738 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
739 ((0 <= available_ix
) &&
740 (!ent
&& pbase_tree_cache
[available_ix
])))
741 available_ix
= my_ix
;
744 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
747 /* Did not find one. Either we got a bogus request or
748 * we need to read and perhaps cache.
750 data
= read_sha1_file(sha1
, type
, &size
);
753 if (strcmp(type
, tree_type
)) {
758 /* We need to either cache or return a throwaway copy */
760 if (available_ix
< 0)
763 ent
= pbase_tree_cache
[available_ix
];
764 my_ix
= available_ix
;
768 nent
= xmalloc(sizeof(*nent
));
769 nent
->temporary
= (available_ix
< 0);
772 /* evict and reuse */
773 free(ent
->tree_data
);
776 hashcpy(nent
->sha1
, sha1
);
777 nent
->tree_data
= data
;
778 nent
->tree_size
= size
;
780 if (!nent
->temporary
)
781 pbase_tree_cache
[my_ix
] = nent
;
785 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
787 if (!cache
->temporary
) {
791 free(cache
->tree_data
);
795 static int name_cmp_len(const char *name
)
798 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
803 static void add_pbase_object(struct tree_desc
*tree
,
806 const char *fullname
)
808 struct name_entry entry
;
810 while (tree_entry(tree
,&entry
)) {
814 if (entry
.pathlen
!= cmplen
||
815 memcmp(entry
.path
, name
, cmplen
) ||
816 !has_sha1_file(entry
.sha1
) ||
817 sha1_object_info(entry
.sha1
, type
, &size
))
819 if (name
[cmplen
] != '/') {
820 unsigned hash
= name_hash(fullname
);
821 add_object_entry(entry
.sha1
, hash
, 1);
824 if (!strcmp(type
, tree_type
)) {
825 struct tree_desc sub
;
826 struct pbase_tree_cache
*tree
;
827 const char *down
= name
+cmplen
+1;
828 int downlen
= name_cmp_len(down
);
830 tree
= pbase_tree_get(entry
.sha1
);
833 sub
.buf
= tree
->tree_data
;
834 sub
.size
= tree
->tree_size
;
836 add_pbase_object(&sub
, down
, downlen
, fullname
);
837 pbase_tree_put(tree
);
842 static unsigned *done_pbase_paths
;
843 static int done_pbase_paths_num
;
844 static int done_pbase_paths_alloc
;
845 static int done_pbase_path_pos(unsigned hash
)
848 int hi
= done_pbase_paths_num
;
850 int mi
= (hi
+ lo
) / 2;
851 if (done_pbase_paths
[mi
] == hash
)
853 if (done_pbase_paths
[mi
] < hash
)
861 static int check_pbase_path(unsigned hash
)
863 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
867 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
868 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
869 done_pbase_paths
= xrealloc(done_pbase_paths
,
870 done_pbase_paths_alloc
*
873 done_pbase_paths_num
++;
874 if (pos
< done_pbase_paths_num
)
875 memmove(done_pbase_paths
+ pos
+ 1,
876 done_pbase_paths
+ pos
,
877 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
878 done_pbase_paths
[pos
] = hash
;
882 static void add_preferred_base_object(const char *name
, unsigned hash
)
884 struct pbase_tree
*it
;
885 int cmplen
= name_cmp_len(name
);
887 if (check_pbase_path(hash
))
890 for (it
= pbase_tree
; it
; it
= it
->next
) {
892 hash
= name_hash("");
893 add_object_entry(it
->pcache
.sha1
, hash
, 1);
896 struct tree_desc tree
;
897 tree
.buf
= it
->pcache
.tree_data
;
898 tree
.size
= it
->pcache
.tree_size
;
899 add_pbase_object(&tree
, name
, cmplen
, name
);
904 static void add_preferred_base(unsigned char *sha1
)
906 struct pbase_tree
*it
;
909 unsigned char tree_sha1
[20];
911 if (window
<= num_preferred_base
++)
914 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
918 for (it
= pbase_tree
; it
; it
= it
->next
) {
919 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
925 it
= xcalloc(1, sizeof(*it
));
926 it
->next
= pbase_tree
;
929 hashcpy(it
->pcache
.sha1
, tree_sha1
);
930 it
->pcache
.tree_data
= data
;
931 it
->pcache
.tree_size
= size
;
934 static void check_object(struct object_entry
*entry
)
938 if (entry
->in_pack
&& !entry
->preferred_base
) {
939 struct packed_git
*p
= entry
->in_pack
;
940 unsigned long left
= p
->pack_size
- entry
->in_pack_offset
;
941 unsigned long size
, used
;
943 struct object_entry
*base_entry
= NULL
;
947 buf
+= entry
->in_pack_offset
;
949 /* We want in_pack_type even if we do not reuse delta.
950 * There is no point not reusing non-delta representations.
952 used
= unpack_object_header_gently(buf
, left
,
953 &entry
->in_pack_type
, &size
);
954 if (!used
|| left
- used
<= 20)
955 die("corrupt pack for %s", sha1_to_hex(entry
->sha1
));
957 /* Check if it is delta, and the base is also an object
958 * we are going to pack. If so we will reuse the existing
961 if (!no_reuse_delta
) {
962 unsigned char c
, *base_name
;
964 /* there is at least 20 bytes left in the pack */
965 switch (entry
->in_pack_type
) {
967 base_name
= buf
+ used
;
975 if (!ofs
|| ofs
& ~(~0UL >> 7))
976 die("delta base offset overflow in pack for %s",
977 sha1_to_hex(entry
->sha1
));
979 ofs
= (ofs
<< 7) + (c
& 127);
981 if (ofs
>= entry
->in_pack_offset
)
982 die("delta base offset out of bound for %s",
983 sha1_to_hex(entry
->sha1
));
984 ofs
= entry
->in_pack_offset
- ofs
;
985 base_name
= find_packed_object_name(p
, ofs
);
991 base_entry
= locate_object_entry(base_name
);
994 entry
->in_pack_header_size
= used
;
998 /* Depth value does not matter - find_deltas()
999 * will never consider reused delta as the
1000 * base object to deltify other objects
1001 * against, in order to avoid circular deltas.
1004 /* uncompressed size of the delta data */
1006 entry
->delta
= base_entry
;
1007 entry
->type
= entry
->in_pack_type
;
1009 entry
->delta_sibling
= base_entry
->delta_child
;
1010 base_entry
->delta_child
= entry
;
1014 /* Otherwise we would do the usual */
1017 if (sha1_object_info(entry
->sha1
, type
, &entry
->size
))
1018 die("unable to get type of object %s",
1019 sha1_to_hex(entry
->sha1
));
1021 if (!strcmp(type
, commit_type
)) {
1022 entry
->type
= OBJ_COMMIT
;
1023 } else if (!strcmp(type
, tree_type
)) {
1024 entry
->type
= OBJ_TREE
;
1025 } else if (!strcmp(type
, blob_type
)) {
1026 entry
->type
= OBJ_BLOB
;
1027 } else if (!strcmp(type
, tag_type
)) {
1028 entry
->type
= OBJ_TAG
;
1030 die("unable to pack object %s of type %s",
1031 sha1_to_hex(entry
->sha1
), type
);
1034 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1036 struct object_entry
*child
= me
->delta_child
;
1039 unsigned int c
= check_delta_limit(child
, n
+ 1);
1042 child
= child
->delta_sibling
;
1047 static void get_object_details(void)
1050 struct object_entry
*entry
;
1053 for (i
= 0, entry
= objects
; i
< nr_objects
; i
++, entry
++)
1054 check_object(entry
);
1056 if (nr_objects
== nr_result
) {
1058 * Depth of objects that depend on the entry -- this
1059 * is subtracted from depth-max to break too deep
1060 * delta chain because of delta data reusing.
1061 * However, we loosen this restriction when we know we
1062 * are creating a thin pack -- it will have to be
1063 * expanded on the other end anyway, so do not
1064 * artificially cut the delta chain and let it go as
1067 for (i
= 0, entry
= objects
; i
< nr_objects
; i
++, entry
++)
1068 if (!entry
->delta
&& entry
->delta_child
)
1069 entry
->delta_limit
=
1070 check_delta_limit(entry
, 1);
1074 typedef int (*entry_sort_t
)(const struct object_entry
*, const struct object_entry
*);
1076 static entry_sort_t current_sort
;
1078 static int sort_comparator(const void *_a
, const void *_b
)
1080 struct object_entry
*a
= *(struct object_entry
**)_a
;
1081 struct object_entry
*b
= *(struct object_entry
**)_b
;
1082 return current_sort(a
,b
);
1085 static struct object_entry
**create_sorted_list(entry_sort_t sort
)
1087 struct object_entry
**list
= xmalloc(nr_objects
* sizeof(struct object_entry
*));
1090 for (i
= 0; i
< nr_objects
; i
++)
1091 list
[i
] = objects
+ i
;
1092 current_sort
= sort
;
1093 qsort(list
, nr_objects
, sizeof(struct object_entry
*), sort_comparator
);
1097 static int sha1_sort(const struct object_entry
*a
, const struct object_entry
*b
)
1099 return hashcmp(a
->sha1
, b
->sha1
);
1102 static struct object_entry
**create_final_object_list(void)
1104 struct object_entry
**list
;
1107 for (i
= nr_result
= 0; i
< nr_objects
; i
++)
1108 if (!objects
[i
].preferred_base
)
1110 list
= xmalloc(nr_result
* sizeof(struct object_entry
*));
1111 for (i
= j
= 0; i
< nr_objects
; i
++) {
1112 if (!objects
[i
].preferred_base
)
1113 list
[j
++] = objects
+ i
;
1115 current_sort
= sha1_sort
;
1116 qsort(list
, nr_result
, sizeof(struct object_entry
*), sort_comparator
);
1120 static int type_size_sort(const struct object_entry
*a
, const struct object_entry
*b
)
1122 if (a
->type
< b
->type
)
1124 if (a
->type
> b
->type
)
1126 if (a
->hash
< b
->hash
)
1128 if (a
->hash
> b
->hash
)
1130 if (a
->preferred_base
< b
->preferred_base
)
1132 if (a
->preferred_base
> b
->preferred_base
)
1134 if (a
->size
< b
->size
)
1136 if (a
->size
> b
->size
)
1138 return a
< b
? -1 : (a
> b
);
1142 struct object_entry
*entry
;
1144 struct delta_index
*index
;
1148 * We search for deltas _backwards_ in a list sorted by type and
1149 * by size, so that we see progressively smaller and smaller files.
1150 * That's because we prefer deltas to be from the bigger file
1151 * to the smaller - deletes are potentially cheaper, but perhaps
1152 * more importantly, the bigger file is likely the more recent
1155 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1158 struct object_entry
*trg_entry
= trg
->entry
;
1159 struct object_entry
*src_entry
= src
->entry
;
1160 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1164 /* Don't bother doing diffs between different types */
1165 if (trg_entry
->type
!= src_entry
->type
)
1168 /* We do not compute delta to *create* objects we are not
1171 if (trg_entry
->preferred_base
)
1175 * We do not bother to try a delta that we discarded
1176 * on an earlier try, but only when reusing delta data.
1178 if (!no_reuse_delta
&& trg_entry
->in_pack
&&
1179 trg_entry
->in_pack
== src_entry
->in_pack
&&
1180 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1181 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1185 * If the current object is at pack edge, take the depth the
1186 * objects that depend on the current object into account --
1187 * otherwise they would become too deep.
1189 if (trg_entry
->delta_child
) {
1190 if (max_depth
<= trg_entry
->delta_limit
)
1192 max_depth
-= trg_entry
->delta_limit
;
1194 if (src_entry
->depth
>= max_depth
)
1197 /* Now some size filtering heuristics. */
1198 trg_size
= trg_entry
->size
;
1199 max_size
= trg_size
/2 - 20;
1200 max_size
= max_size
* (max_depth
- src_entry
->depth
) / max_depth
;
1203 if (trg_entry
->delta
&& trg_entry
->delta_size
<= max_size
)
1204 max_size
= trg_entry
->delta_size
-1;
1205 src_size
= src_entry
->size
;
1206 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1207 if (sizediff
>= max_size
)
1210 /* Load data if not already done */
1212 trg
->data
= read_sha1_file(trg_entry
->sha1
, type
, &sz
);
1214 die("object %s inconsistent object length (%lu vs %lu)",
1215 sha1_to_hex(trg_entry
->sha1
), sz
, trg_size
);
1218 src
->data
= read_sha1_file(src_entry
->sha1
, type
, &sz
);
1220 die("object %s inconsistent object length (%lu vs %lu)",
1221 sha1_to_hex(src_entry
->sha1
), sz
, src_size
);
1224 src
->index
= create_delta_index(src
->data
, src_size
);
1226 die("out of memory");
1229 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1233 trg_entry
->delta
= src_entry
;
1234 trg_entry
->delta_size
= delta_size
;
1235 trg_entry
->depth
= src_entry
->depth
+ 1;
1240 static void progress_interval(int signum
)
1242 progress_update
= 1;
1245 static void find_deltas(struct object_entry
**list
, int window
, int depth
)
1248 unsigned int array_size
= window
* sizeof(struct unpacked
);
1249 struct unpacked
*array
= xmalloc(array_size
);
1250 unsigned processed
= 0;
1251 unsigned last_percent
= 999;
1253 memset(array
, 0, array_size
);
1257 fprintf(stderr
, "Deltifying %d objects.\n", nr_result
);
1260 struct object_entry
*entry
= list
[i
];
1261 struct unpacked
*n
= array
+ idx
;
1264 if (!entry
->preferred_base
)
1268 unsigned percent
= processed
* 100 / nr_result
;
1269 if (percent
!= last_percent
|| progress_update
) {
1270 fprintf(stderr
, "%4u%% (%u/%u) done\r",
1271 percent
, processed
, nr_result
);
1272 progress_update
= 0;
1273 last_percent
= percent
;
1278 /* This happens if we decided to reuse existing
1279 * delta from a pack. "!no_reuse_delta &&" is implied.
1283 if (entry
->size
< 50)
1285 free_delta_index(n
->index
);
1293 unsigned int other_idx
= idx
+ j
;
1295 if (other_idx
>= window
)
1296 other_idx
-= window
;
1297 m
= array
+ other_idx
;
1300 if (try_delta(n
, m
, depth
) < 0)
1303 /* if we made n a delta, and if n is already at max
1304 * depth, leaving it in the window is pointless. we
1305 * should evict it first.
1307 if (entry
->delta
&& depth
<= entry
->depth
)
1316 fputc('\n', stderr
);
1318 for (i
= 0; i
< window
; ++i
) {
1319 free_delta_index(array
[i
].index
);
1320 free(array
[i
].data
);
1325 static void prepare_pack(int window
, int depth
)
1327 get_object_details();
1328 sorted_by_type
= create_sorted_list(type_size_sort
);
1329 if (window
&& depth
)
1330 find_deltas(sorted_by_type
, window
+1, depth
);
1333 static int reuse_cached_pack(unsigned char *sha1
)
1335 static const char cache
[] = "pack-cache/pack-%s.%s";
1336 char *cached_pack
, *cached_idx
;
1337 int ifd
, ofd
, ifd_ix
= -1;
1339 cached_pack
= git_path(cache
, sha1_to_hex(sha1
), "pack");
1340 ifd
= open(cached_pack
, O_RDONLY
);
1344 if (!pack_to_stdout
) {
1345 cached_idx
= git_path(cache
, sha1_to_hex(sha1
), "idx");
1346 ifd_ix
= open(cached_idx
, O_RDONLY
);
1354 fprintf(stderr
, "Reusing %d objects pack %s\n", nr_objects
,
1357 if (pack_to_stdout
) {
1358 if (copy_fd(ifd
, 1))
1363 char name
[PATH_MAX
];
1364 snprintf(name
, sizeof(name
),
1365 "%s-%s.%s", base_name
, sha1_to_hex(sha1
), "pack");
1366 ofd
= open(name
, O_CREAT
| O_EXCL
| O_WRONLY
, 0666);
1368 die("unable to open %s (%s)", name
, strerror(errno
));
1369 if (copy_fd(ifd
, ofd
))
1373 snprintf(name
, sizeof(name
),
1374 "%s-%s.%s", base_name
, sha1_to_hex(sha1
), "idx");
1375 ofd
= open(name
, O_CREAT
| O_EXCL
| O_WRONLY
, 0666);
1377 die("unable to open %s (%s)", name
, strerror(errno
));
1378 if (copy_fd(ifd_ix
, ofd
))
1381 puts(sha1_to_hex(sha1
));
1387 static void setup_progress_signal(void)
1389 struct sigaction sa
;
1392 memset(&sa
, 0, sizeof(sa
));
1393 sa
.sa_handler
= progress_interval
;
1394 sigemptyset(&sa
.sa_mask
);
1395 sa
.sa_flags
= SA_RESTART
;
1396 sigaction(SIGALRM
, &sa
, NULL
);
1398 v
.it_interval
.tv_sec
= 1;
1399 v
.it_interval
.tv_usec
= 0;
1400 v
.it_value
= v
.it_interval
;
1401 setitimer(ITIMER_REAL
, &v
, NULL
);
1404 static int git_pack_config(const char *k
, const char *v
)
1406 if(!strcmp(k
, "pack.window")) {
1407 window
= git_config_int(k
, v
);
1410 return git_default_config(k
, v
);
1413 static void read_object_list_from_stdin(void)
1415 char line
[40 + 1 + PATH_MAX
+ 2];
1416 unsigned char sha1
[20];
1420 if (!fgets(line
, sizeof(line
), stdin
)) {
1424 die("fgets returned NULL, not EOF, not error!");
1426 die("fgets: %s", strerror(errno
));
1430 if (line
[0] == '-') {
1431 if (get_sha1_hex(line
+1, sha1
))
1432 die("expected edge sha1, got garbage:\n %s",
1434 add_preferred_base(sha1
);
1437 if (get_sha1_hex(line
, sha1
))
1438 die("expected sha1, got garbage:\n %s", line
);
1440 hash
= name_hash(line
+41);
1441 add_preferred_base_object(line
+41, hash
);
1442 add_object_entry(sha1
, hash
, 0);
1446 static void show_commit(struct commit
*commit
)
1448 unsigned hash
= name_hash("");
1449 add_preferred_base_object("", hash
);
1450 add_object_entry(commit
->object
.sha1
, hash
, 0);
1453 static void show_object(struct object_array_entry
*p
)
1455 unsigned hash
= name_hash(p
->name
);
1456 add_preferred_base_object(p
->name
, hash
);
1457 add_object_entry(p
->item
->sha1
, hash
, 0);
1460 static void show_edge(struct commit
*commit
)
1462 add_preferred_base(commit
->object
.sha1
);
1465 static void get_object_list(int ac
, const char **av
)
1467 struct rev_info revs
;
1471 init_revisions(&revs
, NULL
);
1472 save_commit_buffer
= 0;
1473 track_object_refs
= 0;
1474 setup_revisions(ac
, av
, &revs
, NULL
);
1476 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
1477 int len
= strlen(line
);
1478 if (line
[len
- 1] == '\n')
1483 if (!strcmp(line
, "--not")) {
1484 flags
^= UNINTERESTING
;
1487 die("not a rev '%s'", line
);
1489 if (handle_revision_arg(line
, &revs
, flags
, 1))
1490 die("bad revision '%s'", line
);
1493 prepare_revision_walk(&revs
);
1494 mark_edges_uninteresting(revs
.commits
, &revs
, show_edge
);
1495 traverse_commit_list(&revs
, show_commit
, show_object
);
1498 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
1502 struct object_entry
**list
;
1503 int use_internal_rev_list
= 0;
1506 const char *rp_av
[64];
1509 rp_av
[0] = "pack-objects";
1510 rp_av
[1] = "--objects"; /* --thin will make it --objects-edge */
1513 git_config(git_pack_config
);
1515 progress
= isatty(2);
1516 for (i
= 1; i
< argc
; i
++) {
1517 const char *arg
= argv
[i
];
1522 if (!strcmp("--non-empty", arg
)) {
1526 if (!strcmp("--local", arg
)) {
1530 if (!strcmp("--incremental", arg
)) {
1534 if (!strncmp("--window=", arg
, 9)) {
1536 window
= strtoul(arg
+9, &end
, 0);
1537 if (!arg
[9] || *end
)
1541 if (!strncmp("--depth=", arg
, 8)) {
1543 depth
= strtoul(arg
+8, &end
, 0);
1544 if (!arg
[8] || *end
)
1548 if (!strcmp("--progress", arg
)) {
1552 if (!strcmp("--all-progress", arg
)) {
1556 if (!strcmp("-q", arg
)) {
1560 if (!strcmp("--no-reuse-delta", arg
)) {
1564 if (!strcmp("--delta-base-offset", arg
)) {
1565 allow_ofs_delta
= 1;
1568 if (!strcmp("--stdout", arg
)) {
1572 if (!strcmp("--revs", arg
)) {
1573 use_internal_rev_list
= 1;
1576 if (!strcmp("--unpacked", arg
) ||
1577 !strncmp("--unpacked=", arg
, 11) ||
1578 !strcmp("--all", arg
)) {
1579 use_internal_rev_list
= 1;
1580 if (ARRAY_SIZE(rp_av
) - 1 <= rp_ac
)
1581 die("too many internal rev-list options");
1582 rp_av
[rp_ac
++] = arg
;
1585 if (!strcmp("--thin", arg
)) {
1586 use_internal_rev_list
= 1;
1588 rp_av
[1] = "--objects-edge";
1594 /* Traditionally "pack-objects [options] base extra" failed;
1595 * we would however want to take refs parameter that would
1596 * have been given to upstream rev-list ourselves, which means
1597 * we somehow want to say what the base name is. So the
1600 * pack-objects [options] base <refs...>
1602 * in other words, we would treat the first non-option as the
1603 * base_name and send everything else to the internal revision
1607 if (!pack_to_stdout
)
1608 base_name
= argv
[i
++];
1610 if (pack_to_stdout
!= !base_name
)
1613 if (!pack_to_stdout
&& thin
)
1614 die("--thin cannot be used to build an indexable pack.");
1616 prepare_packed_git();
1619 fprintf(stderr
, "Generating pack...\n");
1620 setup_progress_signal();
1623 if (!use_internal_rev_list
)
1624 read_object_list_from_stdin();
1626 rp_av
[rp_ac
] = NULL
;
1627 get_object_list(rp_ac
, rp_av
);
1631 fprintf(stderr
, "Done counting %d objects.\n", nr_objects
);
1632 sorted_by_sha
= create_final_object_list();
1633 if (non_empty
&& !nr_result
)
1637 list
= sorted_by_sha
;
1638 for (i
= 0; i
< nr_result
; i
++) {
1639 struct object_entry
*entry
= *list
++;
1640 SHA1_Update(&ctx
, entry
->sha1
, 20);
1642 SHA1_Final(object_list_sha1
, &ctx
);
1643 if (progress
&& (nr_objects
!= nr_result
))
1644 fprintf(stderr
, "Result has %d objects.\n", nr_result
);
1646 if (reuse_cached_pack(object_list_sha1
))
1650 prepare_pack(window
, depth
);
1651 if (progress
== 1 && pack_to_stdout
) {
1652 /* the other end usually displays progress itself */
1653 struct itimerval v
= {{0,},};
1654 setitimer(ITIMER_REAL
, &v
, NULL
);
1655 signal(SIGALRM
, SIG_IGN
);
1656 progress_update
= 0;
1659 if (!pack_to_stdout
) {
1661 puts(sha1_to_hex(object_list_sha1
));
1665 fprintf(stderr
, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1666 nr_result
, written
, written_delta
, reused
, reused_delta
);