11 #include "csum-file.h"
12 #include "tree-walk.h"
15 #include "list-objects.h"
18 #ifdef THREADED_DELTA_SEARCH
22 static const char pack_usage
[] = "\
23 git-pack-objects [{ -q | --progress | --all-progress }] \n\
24 [--max-pack-size=N] [--local] [--incremental] \n\
25 [--window=N] [--window-memory=N] [--depth=N] \n\
26 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
27 [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
28 [--stdout | base-name] [<ref-list | <object-list]";
31 struct pack_idx_entry idx
;
32 unsigned long size
; /* uncompressed size */
33 struct packed_git
*in_pack
; /* already in pack */
35 struct object_entry
*delta
; /* delta base object */
36 struct object_entry
*delta_child
; /* deltified objects who bases me */
37 struct object_entry
*delta_sibling
; /* other deltified objects who
38 * uses the same base as me
40 void *delta_data
; /* cached delta (uncompressed) */
41 unsigned long delta_size
; /* delta data size (uncompressed) */
42 unsigned int hash
; /* name hint hash */
43 enum object_type type
;
44 enum object_type in_pack_type
; /* could be delta */
45 unsigned char in_pack_header_size
;
46 unsigned char preferred_base
; /* we do not pack this, but is available
47 * to be used as the base object to delta
50 unsigned char no_try_delta
;
54 * Objects we are going to pack are collected in objects array (dynamically
55 * expanded). nr_objects & nr_alloc controls this array. They are stored
56 * in the order we see -- typically rev-list --objects order that gives us
57 * nice "minimum seek" order.
59 static struct object_entry
*objects
;
60 static struct object_entry
**written_list
;
61 static uint32_t nr_objects
, nr_alloc
, nr_result
, nr_written
;
64 static int no_reuse_delta
, no_reuse_object
;
66 static int incremental
;
67 static int allow_ofs_delta
;
68 static const char *pack_tmp_name
, *idx_tmp_name
;
69 static char tmpname
[PATH_MAX
];
70 static const char *base_name
;
71 static int progress
= 1;
72 static int window
= 10;
73 static uint32_t pack_size_limit
;
74 static int depth
= 50;
75 static int pack_to_stdout
;
76 static int num_preferred_base
;
77 static struct progress progress_state
;
78 static int pack_compression_level
= Z_DEFAULT_COMPRESSION
;
79 static int pack_compression_seen
;
81 static unsigned long delta_cache_size
= 0;
82 static unsigned long max_delta_cache_size
= 0;
83 static unsigned long cache_max_small_delta_size
= 1000;
85 static unsigned long window_memory_limit
= 0;
88 * The object names in objects array are hashed with this hashtable,
89 * to help looking up the entry by object name.
90 * This hashtable is built after all the objects are seen.
92 static int *object_ix
;
93 static int object_ix_hashsz
;
96 * Pack index for existing packs give us easy access to the offsets into
97 * corresponding pack file where each object's data starts, but the entries
98 * do not store the size of the compressed representation (uncompressed
99 * size is easily available by examining the pack entry header). It is
100 * also rather expensive to find the sha1 for an object given its offset.
102 * We build a hashtable of existing packs (pack_revindex), and keep reverse
103 * index here -- pack index file is sorted by object name mapping to offset;
104 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
105 * ordered by offset, so if you know the offset of an object, next offset
106 * is where its packed representation ends and the index_nr can be used to
107 * get the object sha1 from the main index.
109 struct revindex_entry
{
113 struct pack_revindex
{
114 struct packed_git
*p
;
115 struct revindex_entry
*revindex
;
117 static struct pack_revindex
*pack_revindex
;
118 static int pack_revindex_hashsz
;
123 static uint32_t written
, written_delta
;
124 static uint32_t reused
, reused_delta
;
126 static int pack_revindex_ix(struct packed_git
*p
)
128 unsigned long ui
= (unsigned long)p
;
131 ui
= ui
^ (ui
>> 16); /* defeat structure alignment */
132 i
= (int)(ui
% pack_revindex_hashsz
);
133 while (pack_revindex
[i
].p
) {
134 if (pack_revindex
[i
].p
== p
)
136 if (++i
== pack_revindex_hashsz
)
142 static void prepare_pack_ix(void)
145 struct packed_git
*p
;
146 for (num
= 0, p
= packed_git
; p
; p
= p
->next
)
150 pack_revindex_hashsz
= num
* 11;
151 pack_revindex
= xcalloc(sizeof(*pack_revindex
), pack_revindex_hashsz
);
152 for (p
= packed_git
; p
; p
= p
->next
) {
153 num
= pack_revindex_ix(p
);
155 pack_revindex
[num
].p
= p
;
157 /* revindex elements are lazily initialized */
160 static int cmp_offset(const void *a_
, const void *b_
)
162 const struct revindex_entry
*a
= a_
;
163 const struct revindex_entry
*b
= b_
;
164 return (a
->offset
< b
->offset
) ? -1 : (a
->offset
> b
->offset
) ? 1 : 0;
168 * Ordered list of offsets of objects in the pack.
170 static void prepare_pack_revindex(struct pack_revindex
*rix
)
172 struct packed_git
*p
= rix
->p
;
173 int num_ent
= p
->num_objects
;
175 const char *index
= p
->index_data
;
177 rix
->revindex
= xmalloc(sizeof(*rix
->revindex
) * (num_ent
+ 1));
180 if (p
->index_version
> 1) {
181 const uint32_t *off_32
=
182 (uint32_t *)(index
+ 8 + p
->num_objects
* (20 + 4));
183 const uint32_t *off_64
= off_32
+ p
->num_objects
;
184 for (i
= 0; i
< num_ent
; i
++) {
185 uint32_t off
= ntohl(*off_32
++);
186 if (!(off
& 0x80000000)) {
187 rix
->revindex
[i
].offset
= off
;
189 rix
->revindex
[i
].offset
=
190 ((uint64_t)ntohl(*off_64
++)) << 32;
191 rix
->revindex
[i
].offset
|=
194 rix
->revindex
[i
].nr
= i
;
197 for (i
= 0; i
< num_ent
; i
++) {
198 uint32_t hl
= *((uint32_t *)(index
+ 24 * i
));
199 rix
->revindex
[i
].offset
= ntohl(hl
);
200 rix
->revindex
[i
].nr
= i
;
204 /* This knows the pack format -- the 20-byte trailer
205 * follows immediately after the last object data.
207 rix
->revindex
[num_ent
].offset
= p
->pack_size
- 20;
208 rix
->revindex
[num_ent
].nr
= -1;
209 qsort(rix
->revindex
, num_ent
, sizeof(*rix
->revindex
), cmp_offset
);
212 static struct revindex_entry
* find_packed_object(struct packed_git
*p
,
217 struct pack_revindex
*rix
;
218 struct revindex_entry
*revindex
;
219 num
= pack_revindex_ix(p
);
221 die("internal error: pack revindex uninitialized");
222 rix
= &pack_revindex
[num
];
224 prepare_pack_revindex(rix
);
225 revindex
= rix
->revindex
;
227 hi
= p
->num_objects
+ 1;
229 int mi
= (lo
+ hi
) / 2;
230 if (revindex
[mi
].offset
== ofs
) {
231 return revindex
+ mi
;
233 else if (ofs
< revindex
[mi
].offset
)
238 die("internal error: pack revindex corrupt");
241 static const unsigned char *find_packed_object_name(struct packed_git
*p
,
244 struct revindex_entry
*entry
= find_packed_object(p
, ofs
);
245 return nth_packed_object_sha1(p
, entry
->nr
);
248 static void *delta_against(void *buf
, unsigned long size
, struct object_entry
*entry
)
250 unsigned long othersize
, delta_size
;
251 enum object_type type
;
252 void *otherbuf
= read_sha1_file(entry
->delta
->idx
.sha1
, &type
, &othersize
);
256 die("unable to read %s", sha1_to_hex(entry
->delta
->idx
.sha1
));
257 delta_buf
= diff_delta(otherbuf
, othersize
,
258 buf
, size
, &delta_size
, 0);
259 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
260 die("delta size changed");
267 * The per-object header is a pretty dense thing, which is
268 * - first byte: low four bits are "size", then three bits of "type",
269 * and the high bit is "size continues".
270 * - each byte afterwards: low seven bits are size continuation,
271 * with the high bit being "size continues"
273 static int encode_header(enum object_type type
, unsigned long size
, unsigned char *hdr
)
278 if (type
< OBJ_COMMIT
|| type
> OBJ_REF_DELTA
)
279 die("bad type %d", type
);
281 c
= (type
<< 4) | (size
& 15);
294 * we are going to reuse the existing object data as is. make
295 * sure it is not corrupt.
297 static int check_pack_inflate(struct packed_git
*p
,
298 struct pack_window
**w_curs
,
301 unsigned long expect
)
304 unsigned char fakebuf
[4096], *in
;
307 memset(&stream
, 0, sizeof(stream
));
308 inflateInit(&stream
);
310 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
312 stream
.next_out
= fakebuf
;
313 stream
.avail_out
= sizeof(fakebuf
);
314 st
= inflate(&stream
, Z_FINISH
);
315 offset
+= stream
.next_in
- in
;
316 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
318 return (st
== Z_STREAM_END
&&
319 stream
.total_out
== expect
&&
320 stream
.total_in
== len
) ? 0 : -1;
323 static int check_pack_crc(struct packed_git
*p
, struct pack_window
**w_curs
,
324 off_t offset
, off_t len
, unsigned int nr
)
326 const uint32_t *index_crc
;
327 uint32_t data_crc
= crc32(0, Z_NULL
, 0);
331 void *data
= use_pack(p
, w_curs
, offset
, &avail
);
334 data_crc
= crc32(data_crc
, data
, avail
);
339 index_crc
= p
->index_data
;
340 index_crc
+= 2 + 256 + p
->num_objects
* (20/4) + nr
;
342 return data_crc
!= ntohl(*index_crc
);
345 static void copy_pack_data(struct sha1file
*f
,
346 struct packed_git
*p
,
347 struct pack_window
**w_curs
,
355 in
= use_pack(p
, w_curs
, offset
, &avail
);
357 avail
= (unsigned int)len
;
358 sha1write(f
, in
, avail
);
364 static unsigned long write_object(struct sha1file
*f
,
365 struct object_entry
*entry
,
369 enum object_type type
;
371 unsigned char header
[10];
372 unsigned char dheader
[10];
375 enum object_type obj_type
;
377 /* write limit if limited packsize and not first object */
378 unsigned long limit
= pack_size_limit
&& nr_written
?
379 pack_size_limit
- write_offset
: 0;
381 int usable_delta
= !entry
->delta
? 0 :
382 /* yes if unlimited packfile */
383 !pack_size_limit
? 1 :
384 /* no if base written to previous pack */
385 entry
->delta
->idx
.offset
== (off_t
)-1 ? 0 :
386 /* otherwise double-check written to this
387 * pack, like we do below
389 entry
->delta
->idx
.offset
? 1 : 0;
394 obj_type
= entry
->type
;
396 to_reuse
= 0; /* explicit */
397 else if (!entry
->in_pack
)
398 to_reuse
= 0; /* can't reuse what we don't have */
399 else if (obj_type
== OBJ_REF_DELTA
|| obj_type
== OBJ_OFS_DELTA
)
400 /* check_object() decided it for us ... */
401 to_reuse
= usable_delta
;
402 /* ... but pack split may override that */
403 else if (obj_type
!= entry
->in_pack_type
)
404 to_reuse
= 0; /* pack has delta which is unusable */
405 else if (entry
->delta
)
406 to_reuse
= 0; /* we want to pack afresh */
408 to_reuse
= 1; /* we have it in-pack undeltified,
409 * and we do not need to deltify it.
414 unsigned long maxsize
;
417 buf
= read_sha1_file(entry
->idx
.sha1
, &obj_type
, &size
);
419 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
420 } else if (entry
->delta_data
) {
421 size
= entry
->delta_size
;
422 buf
= entry
->delta_data
;
423 entry
->delta_data
= NULL
;
424 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
425 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
427 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
429 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
430 buf
= delta_against(buf
, size
, entry
);
431 size
= entry
->delta_size
;
432 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
433 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
435 /* compress the data to store and put compressed length in datalen */
436 memset(&stream
, 0, sizeof(stream
));
437 deflateInit(&stream
, pack_compression_level
);
438 maxsize
= deflateBound(&stream
, size
);
439 out
= xmalloc(maxsize
);
441 stream
.next_in
= buf
;
442 stream
.avail_in
= size
;
443 stream
.next_out
= out
;
444 stream
.avail_out
= maxsize
;
445 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
448 datalen
= stream
.total_out
;
451 * The object header is a byte of 'type' followed by zero or
452 * more bytes of length.
454 hdrlen
= encode_header(obj_type
, size
, header
);
456 if (obj_type
== OBJ_OFS_DELTA
) {
458 * Deltas with relative base contain an additional
459 * encoding of the relative offset for the delta
460 * base from this object's position in the pack.
462 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
463 unsigned pos
= sizeof(dheader
) - 1;
464 dheader
[pos
] = ofs
& 127;
466 dheader
[--pos
] = 128 | (--ofs
& 127);
467 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
472 sha1write(f
, header
, hdrlen
);
473 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
474 hdrlen
+= sizeof(dheader
) - pos
;
475 } else if (obj_type
== OBJ_REF_DELTA
) {
477 * Deltas with a base reference contain
478 * an additional 20 bytes for the base sha1.
480 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
485 sha1write(f
, header
, hdrlen
);
486 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
489 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
494 sha1write(f
, header
, hdrlen
);
496 sha1write(f
, out
, datalen
);
501 struct packed_git
*p
= entry
->in_pack
;
502 struct pack_window
*w_curs
= NULL
;
503 struct revindex_entry
*revidx
;
507 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
508 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
511 hdrlen
= encode_header(obj_type
, entry
->size
, header
);
512 offset
= entry
->in_pack_offset
;
513 revidx
= find_packed_object(p
, offset
);
514 datalen
= revidx
[1].offset
- offset
;
515 if (!pack_to_stdout
&& p
->index_version
> 1 &&
516 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
))
517 die("bad packed object CRC for %s", sha1_to_hex(entry
->idx
.sha1
));
518 offset
+= entry
->in_pack_header_size
;
519 datalen
-= entry
->in_pack_header_size
;
520 if (obj_type
== OBJ_OFS_DELTA
) {
521 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
522 unsigned pos
= sizeof(dheader
) - 1;
523 dheader
[pos
] = ofs
& 127;
525 dheader
[--pos
] = 128 | (--ofs
& 127);
526 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
)
528 sha1write(f
, header
, hdrlen
);
529 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
530 hdrlen
+= sizeof(dheader
) - pos
;
531 } else if (obj_type
== OBJ_REF_DELTA
) {
532 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
)
534 sha1write(f
, header
, hdrlen
);
535 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
538 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
)
540 sha1write(f
, header
, hdrlen
);
543 if (!pack_to_stdout
&& p
->index_version
== 1 &&
544 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry
->size
))
545 die("corrupt packed object for %s", sha1_to_hex(entry
->idx
.sha1
));
546 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
554 entry
->idx
.crc32
= crc32_end(f
);
555 return hdrlen
+ datalen
;
558 static off_t
write_one(struct sha1file
*f
,
559 struct object_entry
*e
,
564 /* offset is non zero if object is written already. */
565 if (e
->idx
.offset
|| e
->preferred_base
)
568 /* if we are deltified, write out base object first. */
570 offset
= write_one(f
, e
->delta
, offset
);
575 e
->idx
.offset
= offset
;
576 size
= write_object(f
, e
, offset
);
581 written_list
[nr_written
++] = e
;
583 /* make sure off_t is sufficiently large not to wrap */
584 if (offset
> offset
+ size
)
585 die("pack too large for current definition of off_t");
586 return offset
+ size
;
589 static int open_object_dir_tmp(const char *path
)
591 snprintf(tmpname
, sizeof(tmpname
), "%s/%s", get_object_directory(), path
);
592 return xmkstemp(tmpname
);
595 /* forward declaration for write_pack_file */
596 static int adjust_perm(const char *path
, mode_t mode
);
598 static void write_pack_file(void)
602 off_t offset
, offset_one
, last_obj_offset
= 0;
603 struct pack_header hdr
;
604 int do_progress
= progress
>> pack_to_stdout
;
605 uint32_t nr_remaining
= nr_result
;
608 start_progress(&progress_state
, "Writing %u objects...", "", nr_result
);
609 written_list
= xmalloc(nr_objects
* sizeof(struct object_entry
*));
612 unsigned char sha1
[20];
614 if (pack_to_stdout
) {
615 f
= sha1fd(1, "<stdout>");
617 int fd
= open_object_dir_tmp("tmp_pack_XXXXXX");
618 pack_tmp_name
= xstrdup(tmpname
);
619 f
= sha1fd(fd
, pack_tmp_name
);
622 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
623 hdr
.hdr_version
= htonl(PACK_VERSION
);
624 hdr
.hdr_entries
= htonl(nr_remaining
);
625 sha1write(f
, &hdr
, sizeof(hdr
));
626 offset
= sizeof(hdr
);
628 for (; i
< nr_objects
; i
++) {
629 last_obj_offset
= offset
;
630 offset_one
= write_one(f
, objects
+ i
, offset
);
635 display_progress(&progress_state
, written
);
639 * Did we write the wrong # entries in the header?
640 * If so, rewrite it like in fast-import
642 if (pack_to_stdout
|| nr_written
== nr_remaining
) {
643 sha1close(f
, sha1
, 1);
645 sha1close(f
, sha1
, 0);
646 fixup_pack_header_footer(f
->fd
, sha1
, pack_tmp_name
, nr_written
);
650 if (!pack_to_stdout
) {
651 mode_t mode
= umask(0);
656 idx_tmp_name
= write_idx_file(NULL
,
657 (struct pack_idx_entry
**) written_list
, nr_written
, sha1
);
658 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.pack",
659 base_name
, sha1_to_hex(sha1
));
660 if (adjust_perm(pack_tmp_name
, mode
))
661 die("unable to make temporary pack file readable: %s",
663 if (rename(pack_tmp_name
, tmpname
))
664 die("unable to rename temporary pack file: %s",
666 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.idx",
667 base_name
, sha1_to_hex(sha1
));
668 if (adjust_perm(idx_tmp_name
, mode
))
669 die("unable to make temporary index file readable: %s",
671 if (rename(idx_tmp_name
, tmpname
))
672 die("unable to rename temporary index file: %s",
674 puts(sha1_to_hex(sha1
));
677 /* mark written objects as written to previous pack */
678 for (j
= 0; j
< nr_written
; j
++) {
679 written_list
[j
]->idx
.offset
= (off_t
)-1;
681 nr_remaining
-= nr_written
;
682 } while (nr_remaining
&& i
< nr_objects
);
686 stop_progress(&progress_state
);
687 if (written
!= nr_result
)
688 die("wrote %u objects while expecting %u", written
, nr_result
);
690 * We have scanned through [0 ... i). Since we have written
691 * the correct number of objects, the remaining [i ... nr_objects)
692 * items must be either already written (due to out-of-order delta base)
693 * or a preferred base. Count those which are neither and complain if any.
695 for (j
= 0; i
< nr_objects
; i
++) {
696 struct object_entry
*e
= objects
+ i
;
697 j
+= !e
->idx
.offset
&& !e
->preferred_base
;
700 die("wrote %u objects as expected but %u unwritten", written
, j
);
703 static int locate_object_entry_hash(const unsigned char *sha1
)
707 memcpy(&ui
, sha1
, sizeof(unsigned int));
708 i
= ui
% object_ix_hashsz
;
709 while (0 < object_ix
[i
]) {
710 if (!hashcmp(sha1
, objects
[object_ix
[i
] - 1].idx
.sha1
))
712 if (++i
== object_ix_hashsz
)
718 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
722 if (!object_ix_hashsz
)
725 i
= locate_object_entry_hash(sha1
);
727 return &objects
[object_ix
[i
]-1];
731 static void rehash_objects(void)
734 struct object_entry
*oe
;
736 object_ix_hashsz
= nr_objects
* 3;
737 if (object_ix_hashsz
< 1024)
738 object_ix_hashsz
= 1024;
739 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
740 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
741 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
742 int ix
= locate_object_entry_hash(oe
->idx
.sha1
);
746 object_ix
[ix
] = i
+ 1;
750 static unsigned name_hash(const char *name
)
759 * This effectively just creates a sortable number from the
760 * last sixteen non-whitespace characters. Last characters
761 * count "most", so things that end in ".c" sort together.
763 while ((c
= *name
++) != 0) {
766 hash
= (hash
>> 2) + (c
<< 24);
771 static void setup_delta_attr_check(struct git_attr_check
*check
)
773 static struct git_attr
*attr_delta
;
776 attr_delta
= git_attr("delta", 5);
778 check
[0].attr
= attr_delta
;
781 static int no_try_delta(const char *path
)
783 struct git_attr_check check
[1];
785 setup_delta_attr_check(check
);
786 if (git_checkattr(path
, ARRAY_SIZE(check
), check
))
788 if (ATTR_FALSE(check
->value
))
793 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
794 const char *name
, int exclude
)
796 struct object_entry
*entry
;
797 struct packed_git
*p
, *found_pack
= NULL
;
798 off_t found_offset
= 0;
800 unsigned hash
= name_hash(name
);
802 ix
= nr_objects
? locate_object_entry_hash(sha1
) : -1;
805 entry
= objects
+ object_ix
[ix
] - 1;
806 if (!entry
->preferred_base
)
808 entry
->preferred_base
= 1;
813 for (p
= packed_git
; p
; p
= p
->next
) {
814 off_t offset
= find_pack_entry_one(sha1
, p
);
817 found_offset
= offset
;
824 if (local
&& !p
->pack_local
)
829 if (nr_objects
>= nr_alloc
) {
830 nr_alloc
= (nr_alloc
+ 1024) * 3 / 2;
831 objects
= xrealloc(objects
, nr_alloc
* sizeof(*entry
));
834 entry
= objects
+ nr_objects
++;
835 memset(entry
, 0, sizeof(*entry
));
836 hashcpy(entry
->idx
.sha1
, sha1
);
841 entry
->preferred_base
= 1;
845 entry
->in_pack
= found_pack
;
846 entry
->in_pack_offset
= found_offset
;
849 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
852 object_ix
[-1 - ix
] = nr_objects
;
855 display_progress(&progress_state
, nr_objects
);
857 if (name
&& no_try_delta(name
))
858 entry
->no_try_delta
= 1;
863 struct pbase_tree_cache
{
864 unsigned char sha1
[20];
868 unsigned long tree_size
;
871 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
872 static int pbase_tree_cache_ix(const unsigned char *sha1
)
874 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
876 static int pbase_tree_cache_ix_incr(int ix
)
878 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
881 static struct pbase_tree
{
882 struct pbase_tree
*next
;
883 /* This is a phony "cache" entry; we are not
884 * going to evict it nor find it through _get()
885 * mechanism -- this is for the toplevel node that
886 * would almost always change with any commit.
888 struct pbase_tree_cache pcache
;
891 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
893 struct pbase_tree_cache
*ent
, *nent
;
896 enum object_type type
;
898 int my_ix
= pbase_tree_cache_ix(sha1
);
899 int available_ix
= -1;
901 /* pbase-tree-cache acts as a limited hashtable.
902 * your object will be found at your index or within a few
903 * slots after that slot if it is cached.
905 for (neigh
= 0; neigh
< 8; neigh
++) {
906 ent
= pbase_tree_cache
[my_ix
];
907 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
911 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
912 ((0 <= available_ix
) &&
913 (!ent
&& pbase_tree_cache
[available_ix
])))
914 available_ix
= my_ix
;
917 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
920 /* Did not find one. Either we got a bogus request or
921 * we need to read and perhaps cache.
923 data
= read_sha1_file(sha1
, &type
, &size
);
926 if (type
!= OBJ_TREE
) {
931 /* We need to either cache or return a throwaway copy */
933 if (available_ix
< 0)
936 ent
= pbase_tree_cache
[available_ix
];
937 my_ix
= available_ix
;
941 nent
= xmalloc(sizeof(*nent
));
942 nent
->temporary
= (available_ix
< 0);
945 /* evict and reuse */
946 free(ent
->tree_data
);
949 hashcpy(nent
->sha1
, sha1
);
950 nent
->tree_data
= data
;
951 nent
->tree_size
= size
;
953 if (!nent
->temporary
)
954 pbase_tree_cache
[my_ix
] = nent
;
958 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
960 if (!cache
->temporary
) {
964 free(cache
->tree_data
);
968 static int name_cmp_len(const char *name
)
971 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
976 static void add_pbase_object(struct tree_desc
*tree
,
979 const char *fullname
)
981 struct name_entry entry
;
984 while (tree_entry(tree
,&entry
)) {
985 if (S_ISGITLINK(entry
.mode
))
987 cmp
= tree_entry_len(entry
.path
, entry
.sha1
) != cmplen
? 1 :
988 memcmp(name
, entry
.path
, cmplen
);
993 if (name
[cmplen
] != '/') {
994 add_object_entry(entry
.sha1
,
995 S_ISDIR(entry
.mode
) ? OBJ_TREE
: OBJ_BLOB
,
999 if (S_ISDIR(entry
.mode
)) {
1000 struct tree_desc sub
;
1001 struct pbase_tree_cache
*tree
;
1002 const char *down
= name
+cmplen
+1;
1003 int downlen
= name_cmp_len(down
);
1005 tree
= pbase_tree_get(entry
.sha1
);
1008 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1010 add_pbase_object(&sub
, down
, downlen
, fullname
);
1011 pbase_tree_put(tree
);
1016 static unsigned *done_pbase_paths
;
1017 static int done_pbase_paths_num
;
1018 static int done_pbase_paths_alloc
;
1019 static int done_pbase_path_pos(unsigned hash
)
1022 int hi
= done_pbase_paths_num
;
1024 int mi
= (hi
+ lo
) / 2;
1025 if (done_pbase_paths
[mi
] == hash
)
1027 if (done_pbase_paths
[mi
] < hash
)
1035 static int check_pbase_path(unsigned hash
)
1037 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
1041 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
1042 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
1043 done_pbase_paths
= xrealloc(done_pbase_paths
,
1044 done_pbase_paths_alloc
*
1047 done_pbase_paths_num
++;
1048 if (pos
< done_pbase_paths_num
)
1049 memmove(done_pbase_paths
+ pos
+ 1,
1050 done_pbase_paths
+ pos
,
1051 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
1052 done_pbase_paths
[pos
] = hash
;
1056 static void add_preferred_base_object(const char *name
)
1058 struct pbase_tree
*it
;
1060 unsigned hash
= name_hash(name
);
1062 if (!num_preferred_base
|| check_pbase_path(hash
))
1065 cmplen
= name_cmp_len(name
);
1066 for (it
= pbase_tree
; it
; it
= it
->next
) {
1068 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
1071 struct tree_desc tree
;
1072 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1073 add_pbase_object(&tree
, name
, cmplen
, name
);
1078 static void add_preferred_base(unsigned char *sha1
)
1080 struct pbase_tree
*it
;
1083 unsigned char tree_sha1
[20];
1085 if (window
<= num_preferred_base
++)
1088 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
1092 for (it
= pbase_tree
; it
; it
= it
->next
) {
1093 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
1099 it
= xcalloc(1, sizeof(*it
));
1100 it
->next
= pbase_tree
;
1103 hashcpy(it
->pcache
.sha1
, tree_sha1
);
1104 it
->pcache
.tree_data
= data
;
1105 it
->pcache
.tree_size
= size
;
1108 static void check_object(struct object_entry
*entry
)
1110 if (entry
->in_pack
) {
1111 struct packed_git
*p
= entry
->in_pack
;
1112 struct pack_window
*w_curs
= NULL
;
1113 const unsigned char *base_ref
= NULL
;
1114 struct object_entry
*base_entry
;
1115 unsigned long used
, used_0
;
1118 unsigned char *buf
, c
;
1120 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1123 * We want in_pack_type even if we do not reuse delta
1124 * since non-delta representations could still be reused.
1126 used
= unpack_object_header_gently(buf
, avail
,
1127 &entry
->in_pack_type
,
1131 * Determine if this is a delta and if so whether we can
1132 * reuse it or not. Otherwise let's find out as cheaply as
1133 * possible what the actual type and size for this object is.
1135 switch (entry
->in_pack_type
) {
1137 /* Not a delta hence we've already got all we need. */
1138 entry
->type
= entry
->in_pack_type
;
1139 entry
->in_pack_header_size
= used
;
1140 unuse_pack(&w_curs
);
1143 if (!no_reuse_delta
&& !entry
->preferred_base
)
1144 base_ref
= use_pack(p
, &w_curs
,
1145 entry
->in_pack_offset
+ used
, NULL
);
1146 entry
->in_pack_header_size
= used
+ 20;
1149 buf
= use_pack(p
, &w_curs
,
1150 entry
->in_pack_offset
+ used
, NULL
);
1156 if (!ofs
|| MSB(ofs
, 7))
1157 die("delta base offset overflow in pack for %s",
1158 sha1_to_hex(entry
->idx
.sha1
));
1160 ofs
= (ofs
<< 7) + (c
& 127);
1162 if (ofs
>= entry
->in_pack_offset
)
1163 die("delta base offset out of bound for %s",
1164 sha1_to_hex(entry
->idx
.sha1
));
1165 ofs
= entry
->in_pack_offset
- ofs
;
1166 if (!no_reuse_delta
&& !entry
->preferred_base
)
1167 base_ref
= find_packed_object_name(p
, ofs
);
1168 entry
->in_pack_header_size
= used
+ used_0
;
1172 if (base_ref
&& (base_entry
= locate_object_entry(base_ref
))) {
1174 * If base_ref was set above that means we wish to
1175 * reuse delta data, and we even found that base
1176 * in the list of objects we want to pack. Goodie!
1178 * Depth value does not matter - find_deltas() will
1179 * never consider reused delta as the base object to
1180 * deltify other objects against, in order to avoid
1183 entry
->type
= entry
->in_pack_type
;
1184 entry
->delta
= base_entry
;
1185 entry
->delta_sibling
= base_entry
->delta_child
;
1186 base_entry
->delta_child
= entry
;
1187 unuse_pack(&w_curs
);
1193 * This must be a delta and we already know what the
1194 * final object type is. Let's extract the actual
1195 * object size from the delta header.
1197 entry
->size
= get_size_from_delta(p
, &w_curs
,
1198 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1199 unuse_pack(&w_curs
);
1204 * No choice but to fall back to the recursive delta walk
1205 * with sha1_object_info() to find about the object type
1208 unuse_pack(&w_curs
);
1211 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1212 if (entry
->type
< 0)
1213 die("unable to get type of object %s",
1214 sha1_to_hex(entry
->idx
.sha1
));
1217 static int pack_offset_sort(const void *_a
, const void *_b
)
1219 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1220 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1222 /* avoid filesystem trashing with loose objects */
1223 if (!a
->in_pack
&& !b
->in_pack
)
1224 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1226 if (a
->in_pack
< b
->in_pack
)
1228 if (a
->in_pack
> b
->in_pack
)
1230 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1231 (a
->in_pack_offset
> b
->in_pack_offset
);
1234 static void get_object_details(void)
1237 struct object_entry
**sorted_by_offset
;
1239 sorted_by_offset
= xcalloc(nr_objects
, sizeof(struct object_entry
*));
1240 for (i
= 0; i
< nr_objects
; i
++)
1241 sorted_by_offset
[i
] = objects
+ i
;
1242 qsort(sorted_by_offset
, nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1245 for (i
= 0; i
< nr_objects
; i
++)
1246 check_object(sorted_by_offset
[i
]);
1247 free(sorted_by_offset
);
1250 static int type_size_sort(const void *_a
, const void *_b
)
1252 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1253 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1255 if (a
->type
< b
->type
)
1257 if (a
->type
> b
->type
)
1259 if (a
->hash
< b
->hash
)
1261 if (a
->hash
> b
->hash
)
1263 if (a
->preferred_base
< b
->preferred_base
)
1265 if (a
->preferred_base
> b
->preferred_base
)
1267 if (a
->size
< b
->size
)
1269 if (a
->size
> b
->size
)
1271 return a
> b
? -1 : (a
< b
); /* newest last */
1275 struct object_entry
*entry
;
1277 struct delta_index
*index
;
1281 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1282 unsigned long delta_size
)
1284 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1287 if (delta_size
< cache_max_small_delta_size
)
1290 /* cache delta, if objects are large enough compared to delta size */
1291 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1297 #ifdef THREADED_DELTA_SEARCH
1299 static pthread_mutex_t read_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1300 #define read_lock() pthread_mutex_lock(&read_mutex)
1301 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1303 static pthread_mutex_t progress_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1304 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1305 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1309 #define read_lock() 0
1310 #define read_unlock() 0
1311 #define progress_lock() 0
1312 #define progress_unlock() 0
1317 * We search for deltas _backwards_ in a list sorted by type and
1318 * by size, so that we see progressively smaller and smaller files.
1319 * That's because we prefer deltas to be from the bigger file
1320 * to the smaller - deletes are potentially cheaper, but perhaps
1321 * more importantly, the bigger file is likely the more recent
1324 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1325 unsigned max_depth
, unsigned long *mem_usage
)
1327 struct object_entry
*trg_entry
= trg
->entry
;
1328 struct object_entry
*src_entry
= src
->entry
;
1329 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1331 enum object_type type
;
1334 /* Don't bother doing diffs between different types */
1335 if (trg_entry
->type
!= src_entry
->type
)
1339 * We do not bother to try a delta that we discarded
1340 * on an earlier try, but only when reusing delta data.
1342 if (!no_reuse_delta
&& trg_entry
->in_pack
&&
1343 trg_entry
->in_pack
== src_entry
->in_pack
&&
1344 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1345 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1348 /* Let's not bust the allowed depth. */
1349 if (src
->depth
>= max_depth
)
1352 /* Now some size filtering heuristics. */
1353 trg_size
= trg_entry
->size
;
1354 if (!trg_entry
->delta
) {
1355 max_size
= trg_size
/2 - 20;
1358 max_size
= trg_entry
->delta_size
;
1359 ref_depth
= trg
->depth
;
1361 max_size
= max_size
* (max_depth
- src
->depth
) /
1362 (max_depth
- ref_depth
+ 1);
1365 src_size
= src_entry
->size
;
1366 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1367 if (sizediff
>= max_size
)
1369 if (trg_size
< src_size
/ 32)
1372 /* Load data if not already done */
1375 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1378 die("object %s cannot be read",
1379 sha1_to_hex(trg_entry
->idx
.sha1
));
1381 die("object %s inconsistent object length (%lu vs %lu)",
1382 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1387 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1390 die("object %s cannot be read",
1391 sha1_to_hex(src_entry
->idx
.sha1
));
1393 die("object %s inconsistent object length (%lu vs %lu)",
1394 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1398 src
->index
= create_delta_index(src
->data
, src_size
);
1400 static int warned
= 0;
1402 warning("suboptimal pack - out of memory");
1405 *mem_usage
+= sizeof_delta_index(src
->index
);
1408 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1412 if (trg_entry
->delta
) {
1413 /* Prefer only shallower same-sized deltas. */
1414 if (delta_size
== trg_entry
->delta_size
&&
1415 src
->depth
+ 1 >= trg
->depth
) {
1421 trg_entry
->delta
= src_entry
;
1422 trg_entry
->delta_size
= delta_size
;
1423 trg
->depth
= src
->depth
+ 1;
1425 if (trg_entry
->delta_data
) {
1426 delta_cache_size
-= trg_entry
->delta_size
;
1427 free(trg_entry
->delta_data
);
1428 trg_entry
->delta_data
= NULL
;
1431 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1432 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1433 delta_cache_size
+= trg_entry
->delta_size
;
1439 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1441 struct object_entry
*child
= me
->delta_child
;
1444 unsigned int c
= check_delta_limit(child
, n
+ 1);
1447 child
= child
->delta_sibling
;
1452 static unsigned long free_unpacked(struct unpacked
*n
)
1454 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1455 free_delta_index(n
->index
);
1458 freed_mem
+= n
->entry
->size
;
1467 static void find_deltas(struct object_entry
**list
, unsigned list_size
,
1468 int window
, int depth
, unsigned *processed
)
1470 uint32_t i
= list_size
, idx
= 0, count
= 0;
1471 unsigned int array_size
= window
* sizeof(struct unpacked
);
1472 struct unpacked
*array
;
1473 unsigned long mem_usage
= 0;
1475 array
= xmalloc(array_size
);
1476 memset(array
, 0, array_size
);
1479 struct object_entry
*entry
= list
[--i
];
1480 struct unpacked
*n
= array
+ idx
;
1481 int j
, max_depth
, best_base
= -1;
1483 mem_usage
-= free_unpacked(n
);
1486 while (window_memory_limit
&&
1487 mem_usage
> window_memory_limit
&&
1489 uint32_t tail
= (idx
+ window
- count
) % window
;
1490 mem_usage
-= free_unpacked(array
+ tail
);
1494 /* We do not compute delta to *create* objects we are not
1497 if (entry
->preferred_base
)
1503 display_progress(&progress_state
, *processed
);
1507 * If the current object is at pack edge, take the depth the
1508 * objects that depend on the current object into account
1509 * otherwise they would become too deep.
1512 if (entry
->delta_child
) {
1513 max_depth
-= check_delta_limit(entry
, 0);
1521 uint32_t other_idx
= idx
+ j
;
1523 if (other_idx
>= window
)
1524 other_idx
-= window
;
1525 m
= array
+ other_idx
;
1528 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1532 best_base
= other_idx
;
1535 /* if we made n a delta, and if n is already at max
1536 * depth, leaving it in the window is pointless. we
1537 * should evict it first.
1539 if (entry
->delta
&& depth
<= n
->depth
)
1543 * Move the best delta base up in the window, after the
1544 * currently deltified object, to keep it longer. It will
1545 * be the first base object to be attempted next.
1548 struct unpacked swap
= array
[best_base
];
1549 int dist
= (window
+ idx
- best_base
) % window
;
1550 int dst
= best_base
;
1552 int src
= (dst
+ 1) % window
;
1553 array
[dst
] = array
[src
];
1561 if (count
+ 1 < window
)
1567 for (i
= 0; i
< window
; ++i
) {
1568 free_delta_index(array
[i
].index
);
1569 free(array
[i
].data
);
1574 #ifdef THREADED_DELTA_SEARCH
1576 struct thread_params
{
1578 struct object_entry
**list
;
1582 unsigned *processed
;
1585 static pthread_mutex_t data_request
= PTHREAD_MUTEX_INITIALIZER
;
1586 static pthread_mutex_t data_ready
= PTHREAD_MUTEX_INITIALIZER
;
1587 static pthread_mutex_t data_provider
= PTHREAD_MUTEX_INITIALIZER
;
1588 static struct thread_params
*data_requester
;
1590 static void *threaded_find_deltas(void *arg
)
1592 struct thread_params
*me
= arg
;
1595 pthread_mutex_lock(&data_request
);
1596 data_requester
= me
;
1597 pthread_mutex_unlock(&data_provider
);
1598 pthread_mutex_lock(&data_ready
);
1603 find_deltas(me
->list
, me
->list_size
,
1604 me
->window
, me
->depth
, me
->processed
);
1608 #define NR_THREADS 4
1610 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
1611 int window
, int depth
, unsigned *processed
)
1613 struct thread_params p
[NR_THREADS
];
1615 unsigned chunk_size
;
1617 pthread_mutex_lock(&data_provider
);
1618 pthread_mutex_lock(&data_ready
);
1620 for (i
= 0; i
< NR_THREADS
; i
++) {
1621 p
[i
].window
= window
;
1623 p
[i
].processed
= processed
;
1624 ret
= pthread_create(&p
[i
].thread
, NULL
,
1625 threaded_find_deltas
, &p
[i
]);
1627 die("unable to create thread: %s", strerror(ret
));
1630 /* this should be auto-tuned somehow */
1631 chunk_size
= window
* 1000;
1634 unsigned sublist_size
= chunk_size
;
1635 if (sublist_size
> list_size
)
1636 sublist_size
= list_size
;
1638 pthread_mutex_lock(&data_provider
);
1639 data_requester
->list
= list
;
1640 data_requester
->list_size
= sublist_size
;
1641 pthread_mutex_unlock(&data_ready
);
1643 list
+= sublist_size
;
1644 list_size
-= sublist_size
;
1645 if (!sublist_size
) {
1646 pthread_join(data_requester
->thread
, NULL
);
1649 pthread_mutex_unlock(&data_request
);
1654 #define ll_find_deltas find_deltas
1657 static void prepare_pack(int window
, int depth
)
1659 struct object_entry
**delta_list
;
1660 uint32_t i
, n
, nr_deltas
;
1662 get_object_details();
1664 if (!nr_objects
|| !window
|| !depth
)
1667 delta_list
= xmalloc(nr_objects
* sizeof(*delta_list
));
1670 for (i
= 0; i
< nr_objects
; i
++) {
1671 struct object_entry
*entry
= objects
+ i
;
1674 /* This happens if we decided to reuse existing
1675 * delta from a pack. "!no_reuse_delta &&" is implied.
1679 if (entry
->size
< 50)
1682 if (entry
->no_try_delta
)
1685 if (!entry
->preferred_base
)
1688 delta_list
[n
++] = entry
;
1692 unsigned nr_done
= 0;
1694 start_progress(&progress_state
,
1695 "Deltifying %u objects...", "",
1697 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
1698 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
1700 stop_progress(&progress_state
);
1701 if (nr_done
!= nr_deltas
)
1702 die("inconsistency with delta count");
1707 static int git_pack_config(const char *k
, const char *v
)
1709 if(!strcmp(k
, "pack.window")) {
1710 window
= git_config_int(k
, v
);
1713 if (!strcmp(k
, "pack.windowmemory")) {
1714 window_memory_limit
= git_config_ulong(k
, v
);
1717 if (!strcmp(k
, "pack.depth")) {
1718 depth
= git_config_int(k
, v
);
1721 if (!strcmp(k
, "pack.compression")) {
1722 int level
= git_config_int(k
, v
);
1724 level
= Z_DEFAULT_COMPRESSION
;
1725 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
1726 die("bad pack compression level %d", level
);
1727 pack_compression_level
= level
;
1728 pack_compression_seen
= 1;
1731 if (!strcmp(k
, "pack.deltacachesize")) {
1732 max_delta_cache_size
= git_config_int(k
, v
);
1735 if (!strcmp(k
, "pack.deltacachelimit")) {
1736 cache_max_small_delta_size
= git_config_int(k
, v
);
1739 return git_default_config(k
, v
);
1742 static void read_object_list_from_stdin(void)
1744 char line
[40 + 1 + PATH_MAX
+ 2];
1745 unsigned char sha1
[20];
1748 if (!fgets(line
, sizeof(line
), stdin
)) {
1752 die("fgets returned NULL, not EOF, not error!");
1754 die("fgets: %s", strerror(errno
));
1758 if (line
[0] == '-') {
1759 if (get_sha1_hex(line
+1, sha1
))
1760 die("expected edge sha1, got garbage:\n %s",
1762 add_preferred_base(sha1
);
1765 if (get_sha1_hex(line
, sha1
))
1766 die("expected sha1, got garbage:\n %s", line
);
1768 add_preferred_base_object(line
+41);
1769 add_object_entry(sha1
, 0, line
+41, 0);
1773 static void show_commit(struct commit
*commit
)
1775 add_object_entry(commit
->object
.sha1
, OBJ_COMMIT
, NULL
, 0);
1778 static void show_object(struct object_array_entry
*p
)
1780 add_preferred_base_object(p
->name
);
1781 add_object_entry(p
->item
->sha1
, p
->item
->type
, p
->name
, 0);
1784 static void show_edge(struct commit
*commit
)
1786 add_preferred_base(commit
->object
.sha1
);
1789 static void get_object_list(int ac
, const char **av
)
1791 struct rev_info revs
;
1795 init_revisions(&revs
, NULL
);
1796 save_commit_buffer
= 0;
1797 track_object_refs
= 0;
1798 setup_revisions(ac
, av
, &revs
, NULL
);
1800 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
1801 int len
= strlen(line
);
1802 if (line
[len
- 1] == '\n')
1807 if (!strcmp(line
, "--not")) {
1808 flags
^= UNINTERESTING
;
1811 die("not a rev '%s'", line
);
1813 if (handle_revision_arg(line
, &revs
, flags
, 1))
1814 die("bad revision '%s'", line
);
1817 prepare_revision_walk(&revs
);
1818 mark_edges_uninteresting(revs
.commits
, &revs
, show_edge
);
1819 traverse_commit_list(&revs
, show_commit
, show_object
);
1822 static int adjust_perm(const char *path
, mode_t mode
)
1824 if (chmod(path
, mode
))
1826 return adjust_shared_perm(path
);
1829 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
1831 int use_internal_rev_list
= 0;
1835 int rp_ac_alloc
= 64;
1838 rp_av
= xcalloc(rp_ac_alloc
, sizeof(*rp_av
));
1840 rp_av
[0] = "pack-objects";
1841 rp_av
[1] = "--objects"; /* --thin will make it --objects-edge */
1844 git_config(git_pack_config
);
1845 if (!pack_compression_seen
&& core_compression_seen
)
1846 pack_compression_level
= core_compression_level
;
1848 progress
= isatty(2);
1849 for (i
= 1; i
< argc
; i
++) {
1850 const char *arg
= argv
[i
];
1855 if (!strcmp("--non-empty", arg
)) {
1859 if (!strcmp("--local", arg
)) {
1863 if (!strcmp("--incremental", arg
)) {
1867 if (!prefixcmp(arg
, "--compression=")) {
1869 int level
= strtoul(arg
+14, &end
, 0);
1870 if (!arg
[14] || *end
)
1873 level
= Z_DEFAULT_COMPRESSION
;
1874 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
1875 die("bad pack compression level %d", level
);
1876 pack_compression_level
= level
;
1879 if (!prefixcmp(arg
, "--max-pack-size=")) {
1881 pack_size_limit
= strtoul(arg
+16, &end
, 0) * 1024 * 1024;
1882 if (!arg
[16] || *end
)
1886 if (!prefixcmp(arg
, "--window=")) {
1888 window
= strtoul(arg
+9, &end
, 0);
1889 if (!arg
[9] || *end
)
1893 if (!prefixcmp(arg
, "--window-memory=")) {
1894 if (!git_parse_ulong(arg
+16, &window_memory_limit
))
1898 if (!prefixcmp(arg
, "--depth=")) {
1900 depth
= strtoul(arg
+8, &end
, 0);
1901 if (!arg
[8] || *end
)
1905 if (!strcmp("--progress", arg
)) {
1909 if (!strcmp("--all-progress", arg
)) {
1913 if (!strcmp("-q", arg
)) {
1917 if (!strcmp("--no-reuse-delta", arg
)) {
1921 if (!strcmp("--no-reuse-object", arg
)) {
1922 no_reuse_object
= no_reuse_delta
= 1;
1925 if (!strcmp("--delta-base-offset", arg
)) {
1926 allow_ofs_delta
= 1;
1929 if (!strcmp("--stdout", arg
)) {
1933 if (!strcmp("--revs", arg
)) {
1934 use_internal_rev_list
= 1;
1937 if (!strcmp("--unpacked", arg
) ||
1938 !prefixcmp(arg
, "--unpacked=") ||
1939 !strcmp("--reflog", arg
) ||
1940 !strcmp("--all", arg
)) {
1941 use_internal_rev_list
= 1;
1942 if (rp_ac
>= rp_ac_alloc
- 1) {
1943 rp_ac_alloc
= alloc_nr(rp_ac_alloc
);
1944 rp_av
= xrealloc(rp_av
,
1945 rp_ac_alloc
* sizeof(*rp_av
));
1947 rp_av
[rp_ac
++] = arg
;
1950 if (!strcmp("--thin", arg
)) {
1951 use_internal_rev_list
= 1;
1953 rp_av
[1] = "--objects-edge";
1956 if (!prefixcmp(arg
, "--index-version=")) {
1958 pack_idx_default_version
= strtoul(arg
+ 16, &c
, 10);
1959 if (pack_idx_default_version
> 2)
1962 pack_idx_off32_limit
= strtoul(c
+1, &c
, 0);
1963 if (*c
|| pack_idx_off32_limit
& 0x80000000)
1970 /* Traditionally "pack-objects [options] base extra" failed;
1971 * we would however want to take refs parameter that would
1972 * have been given to upstream rev-list ourselves, which means
1973 * we somehow want to say what the base name is. So the
1976 * pack-objects [options] base <refs...>
1978 * in other words, we would treat the first non-option as the
1979 * base_name and send everything else to the internal revision
1983 if (!pack_to_stdout
)
1984 base_name
= argv
[i
++];
1986 if (pack_to_stdout
!= !base_name
)
1989 if (pack_to_stdout
&& pack_size_limit
)
1990 die("--max-pack-size cannot be used to build a pack for transfer.");
1992 if (!pack_to_stdout
&& thin
)
1993 die("--thin cannot be used to build an indexable pack.");
1995 prepare_packed_git();
1998 start_progress(&progress_state
, "Generating pack...",
1999 "Counting objects: ", 0);
2000 if (!use_internal_rev_list
)
2001 read_object_list_from_stdin();
2003 rp_av
[rp_ac
] = NULL
;
2004 get_object_list(rp_ac
, rp_av
);
2007 stop_progress(&progress_state
);
2008 fprintf(stderr
, "Done counting %u objects.\n", nr_objects
);
2011 if (non_empty
&& !nr_result
)
2013 if (progress
&& (nr_objects
!= nr_result
))
2014 fprintf(stderr
, "Result has %u objects.\n", nr_result
);
2016 prepare_pack(window
, depth
);
2019 fprintf(stderr
, "Total %u (delta %u), reused %u (delta %u)\n",
2020 written
, written_delta
, reused
, reused_delta
);