11 #include "csum-file.h"
12 #include "tree-walk.h"
15 #include "list-objects.h"
18 #ifdef THREADED_DELTA_SEARCH
19 #include "thread-utils.h"
23 static const char pack_usage
[] = "\
24 git-pack-objects [{ -q | --progress | --all-progress }] \n\
25 [--max-pack-size=N] [--local] [--incremental] \n\
26 [--window=N] [--window-memory=N] [--depth=N] \n\
27 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
28 [--threads=N] [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
29 [--stdout | base-name] [--keep-unreachable] [<ref-list | <object-list]";
32 struct pack_idx_entry idx
;
33 unsigned long size
; /* uncompressed size */
34 struct packed_git
*in_pack
; /* already in pack */
36 struct object_entry
*delta
; /* delta base object */
37 struct object_entry
*delta_child
; /* deltified objects who bases me */
38 struct object_entry
*delta_sibling
; /* other deltified objects who
39 * uses the same base as me
41 void *delta_data
; /* cached delta (uncompressed) */
42 unsigned long delta_size
; /* delta data size (uncompressed) */
43 unsigned int hash
; /* name hint hash */
44 enum object_type type
;
45 enum object_type in_pack_type
; /* could be delta */
46 unsigned char in_pack_header_size
;
47 unsigned char preferred_base
; /* we do not pack this, but is available
48 * to be used as the base object to delta
51 unsigned char no_try_delta
;
55 * Objects we are going to pack are collected in objects array (dynamically
56 * expanded). nr_objects & nr_alloc controls this array. They are stored
57 * in the order we see -- typically rev-list --objects order that gives us
58 * nice "minimum seek" order.
60 static struct object_entry
*objects
;
61 static struct pack_idx_entry
**written_list
;
62 static uint32_t nr_objects
, nr_alloc
, nr_result
, nr_written
;
65 static int no_reuse_delta
, no_reuse_object
, keep_unreachable
;
67 static int incremental
;
68 static int allow_ofs_delta
;
69 static const char *base_name
;
70 static int progress
= 1;
71 static int window
= 10;
72 static uint32_t pack_size_limit
, pack_size_limit_cfg
;
73 static int depth
= 50;
74 static int delta_search_threads
= 1;
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
->idx
;
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 /* forward declaration for write_pack_file */
590 static int adjust_perm(const char *path
, mode_t mode
);
592 static void write_pack_file(void)
596 off_t offset
, offset_one
, last_obj_offset
= 0;
597 struct pack_header hdr
;
598 int do_progress
= progress
>> pack_to_stdout
;
599 uint32_t nr_remaining
= nr_result
;
602 progress_state
= start_progress("Writing objects", nr_result
);
603 written_list
= xmalloc(nr_objects
* sizeof(*written_list
));
606 unsigned char sha1
[20];
607 char *pack_tmp_name
= NULL
;
609 if (pack_to_stdout
) {
610 f
= sha1fd_throughput(1, "<stdout>", progress_state
);
612 char tmpname
[PATH_MAX
];
614 snprintf(tmpname
, sizeof(tmpname
),
615 "%s/tmp_pack_XXXXXX", get_object_directory());
616 fd
= xmkstemp(tmpname
);
617 pack_tmp_name
= xstrdup(tmpname
);
618 f
= sha1fd(fd
, pack_tmp_name
);
621 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
622 hdr
.hdr_version
= htonl(PACK_VERSION
);
623 hdr
.hdr_entries
= htonl(nr_remaining
);
624 sha1write(f
, &hdr
, sizeof(hdr
));
625 offset
= sizeof(hdr
);
627 for (; i
< nr_objects
; i
++) {
628 last_obj_offset
= offset
;
629 offset_one
= write_one(f
, objects
+ i
, offset
);
633 display_progress(progress_state
, written
);
637 * Did we write the wrong # entries in the header?
638 * If so, rewrite it like in fast-import
640 if (pack_to_stdout
|| nr_written
== nr_remaining
) {
641 sha1close(f
, sha1
, 1);
643 int fd
= sha1close(f
, NULL
, 0);
644 fixup_pack_header_footer(fd
, sha1
, pack_tmp_name
, nr_written
);
648 if (!pack_to_stdout
) {
649 mode_t mode
= umask(0);
650 char *idx_tmp_name
, tmpname
[PATH_MAX
];
655 idx_tmp_name
= write_idx_file(NULL
, written_list
,
657 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.pack",
658 base_name
, sha1_to_hex(sha1
));
659 if (adjust_perm(pack_tmp_name
, mode
))
660 die("unable to make temporary pack file readable: %s",
662 if (rename(pack_tmp_name
, tmpname
))
663 die("unable to rename temporary pack file: %s",
665 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.idx",
666 base_name
, sha1_to_hex(sha1
));
667 if (adjust_perm(idx_tmp_name
, mode
))
668 die("unable to make temporary index file readable: %s",
670 if (rename(idx_tmp_name
, tmpname
))
671 die("unable to rename temporary index file: %s",
675 puts(sha1_to_hex(sha1
));
678 /* mark written objects as written to previous pack */
679 for (j
= 0; j
< nr_written
; j
++) {
680 written_list
[j
]->offset
= (off_t
)-1;
682 nr_remaining
-= nr_written
;
683 } 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
;
854 display_progress(progress_state
, nr_objects
);
856 if (name
&& no_try_delta(name
))
857 entry
->no_try_delta
= 1;
862 struct pbase_tree_cache
{
863 unsigned char sha1
[20];
867 unsigned long tree_size
;
870 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
871 static int pbase_tree_cache_ix(const unsigned char *sha1
)
873 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
875 static int pbase_tree_cache_ix_incr(int ix
)
877 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
880 static struct pbase_tree
{
881 struct pbase_tree
*next
;
882 /* This is a phony "cache" entry; we are not
883 * going to evict it nor find it through _get()
884 * mechanism -- this is for the toplevel node that
885 * would almost always change with any commit.
887 struct pbase_tree_cache pcache
;
890 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
892 struct pbase_tree_cache
*ent
, *nent
;
895 enum object_type type
;
897 int my_ix
= pbase_tree_cache_ix(sha1
);
898 int available_ix
= -1;
900 /* pbase-tree-cache acts as a limited hashtable.
901 * your object will be found at your index or within a few
902 * slots after that slot if it is cached.
904 for (neigh
= 0; neigh
< 8; neigh
++) {
905 ent
= pbase_tree_cache
[my_ix
];
906 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
910 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
911 ((0 <= available_ix
) &&
912 (!ent
&& pbase_tree_cache
[available_ix
])))
913 available_ix
= my_ix
;
916 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
919 /* Did not find one. Either we got a bogus request or
920 * we need to read and perhaps cache.
922 data
= read_sha1_file(sha1
, &type
, &size
);
925 if (type
!= OBJ_TREE
) {
930 /* We need to either cache or return a throwaway copy */
932 if (available_ix
< 0)
935 ent
= pbase_tree_cache
[available_ix
];
936 my_ix
= available_ix
;
940 nent
= xmalloc(sizeof(*nent
));
941 nent
->temporary
= (available_ix
< 0);
944 /* evict and reuse */
945 free(ent
->tree_data
);
948 hashcpy(nent
->sha1
, sha1
);
949 nent
->tree_data
= data
;
950 nent
->tree_size
= size
;
952 if (!nent
->temporary
)
953 pbase_tree_cache
[my_ix
] = nent
;
957 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
959 if (!cache
->temporary
) {
963 free(cache
->tree_data
);
967 static int name_cmp_len(const char *name
)
970 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
975 static void add_pbase_object(struct tree_desc
*tree
,
978 const char *fullname
)
980 struct name_entry entry
;
983 while (tree_entry(tree
,&entry
)) {
984 if (S_ISGITLINK(entry
.mode
))
986 cmp
= tree_entry_len(entry
.path
, entry
.sha1
) != cmplen
? 1 :
987 memcmp(name
, entry
.path
, cmplen
);
992 if (name
[cmplen
] != '/') {
993 add_object_entry(entry
.sha1
,
994 object_type(entry
.mode
),
998 if (S_ISDIR(entry
.mode
)) {
999 struct tree_desc sub
;
1000 struct pbase_tree_cache
*tree
;
1001 const char *down
= name
+cmplen
+1;
1002 int downlen
= name_cmp_len(down
);
1004 tree
= pbase_tree_get(entry
.sha1
);
1007 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1009 add_pbase_object(&sub
, down
, downlen
, fullname
);
1010 pbase_tree_put(tree
);
1015 static unsigned *done_pbase_paths
;
1016 static int done_pbase_paths_num
;
1017 static int done_pbase_paths_alloc
;
1018 static int done_pbase_path_pos(unsigned hash
)
1021 int hi
= done_pbase_paths_num
;
1023 int mi
= (hi
+ lo
) / 2;
1024 if (done_pbase_paths
[mi
] == hash
)
1026 if (done_pbase_paths
[mi
] < hash
)
1034 static int check_pbase_path(unsigned hash
)
1036 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
1040 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
1041 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
1042 done_pbase_paths
= xrealloc(done_pbase_paths
,
1043 done_pbase_paths_alloc
*
1046 done_pbase_paths_num
++;
1047 if (pos
< done_pbase_paths_num
)
1048 memmove(done_pbase_paths
+ pos
+ 1,
1049 done_pbase_paths
+ pos
,
1050 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
1051 done_pbase_paths
[pos
] = hash
;
1055 static void add_preferred_base_object(const char *name
)
1057 struct pbase_tree
*it
;
1059 unsigned hash
= name_hash(name
);
1061 if (!num_preferred_base
|| check_pbase_path(hash
))
1064 cmplen
= name_cmp_len(name
);
1065 for (it
= pbase_tree
; it
; it
= it
->next
) {
1067 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
1070 struct tree_desc tree
;
1071 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1072 add_pbase_object(&tree
, name
, cmplen
, name
);
1077 static void add_preferred_base(unsigned char *sha1
)
1079 struct pbase_tree
*it
;
1082 unsigned char tree_sha1
[20];
1084 if (window
<= num_preferred_base
++)
1087 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
1091 for (it
= pbase_tree
; it
; it
= it
->next
) {
1092 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
1098 it
= xcalloc(1, sizeof(*it
));
1099 it
->next
= pbase_tree
;
1102 hashcpy(it
->pcache
.sha1
, tree_sha1
);
1103 it
->pcache
.tree_data
= data
;
1104 it
->pcache
.tree_size
= size
;
1107 static void check_object(struct object_entry
*entry
)
1109 if (entry
->in_pack
) {
1110 struct packed_git
*p
= entry
->in_pack
;
1111 struct pack_window
*w_curs
= NULL
;
1112 const unsigned char *base_ref
= NULL
;
1113 struct object_entry
*base_entry
;
1114 unsigned long used
, used_0
;
1117 unsigned char *buf
, c
;
1119 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1122 * We want in_pack_type even if we do not reuse delta
1123 * since non-delta representations could still be reused.
1125 used
= unpack_object_header_gently(buf
, avail
,
1126 &entry
->in_pack_type
,
1130 * Determine if this is a delta and if so whether we can
1131 * reuse it or not. Otherwise let's find out as cheaply as
1132 * possible what the actual type and size for this object is.
1134 switch (entry
->in_pack_type
) {
1136 /* Not a delta hence we've already got all we need. */
1137 entry
->type
= entry
->in_pack_type
;
1138 entry
->in_pack_header_size
= used
;
1139 unuse_pack(&w_curs
);
1142 if (!no_reuse_delta
&& !entry
->preferred_base
)
1143 base_ref
= use_pack(p
, &w_curs
,
1144 entry
->in_pack_offset
+ used
, NULL
);
1145 entry
->in_pack_header_size
= used
+ 20;
1148 buf
= use_pack(p
, &w_curs
,
1149 entry
->in_pack_offset
+ used
, NULL
);
1155 if (!ofs
|| MSB(ofs
, 7))
1156 die("delta base offset overflow in pack for %s",
1157 sha1_to_hex(entry
->idx
.sha1
));
1159 ofs
= (ofs
<< 7) + (c
& 127);
1161 if (ofs
>= entry
->in_pack_offset
)
1162 die("delta base offset out of bound for %s",
1163 sha1_to_hex(entry
->idx
.sha1
));
1164 ofs
= entry
->in_pack_offset
- ofs
;
1165 if (!no_reuse_delta
&& !entry
->preferred_base
)
1166 base_ref
= find_packed_object_name(p
, ofs
);
1167 entry
->in_pack_header_size
= used
+ used_0
;
1171 if (base_ref
&& (base_entry
= locate_object_entry(base_ref
))) {
1173 * If base_ref was set above that means we wish to
1174 * reuse delta data, and we even found that base
1175 * in the list of objects we want to pack. Goodie!
1177 * Depth value does not matter - find_deltas() will
1178 * never consider reused delta as the base object to
1179 * deltify other objects against, in order to avoid
1182 entry
->type
= entry
->in_pack_type
;
1183 entry
->delta
= base_entry
;
1184 entry
->delta_sibling
= base_entry
->delta_child
;
1185 base_entry
->delta_child
= entry
;
1186 unuse_pack(&w_curs
);
1192 * This must be a delta and we already know what the
1193 * final object type is. Let's extract the actual
1194 * object size from the delta header.
1196 entry
->size
= get_size_from_delta(p
, &w_curs
,
1197 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1198 unuse_pack(&w_curs
);
1203 * No choice but to fall back to the recursive delta walk
1204 * with sha1_object_info() to find about the object type
1207 unuse_pack(&w_curs
);
1210 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1211 if (entry
->type
< 0)
1212 die("unable to get type of object %s",
1213 sha1_to_hex(entry
->idx
.sha1
));
1216 static int pack_offset_sort(const void *_a
, const void *_b
)
1218 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1219 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1221 /* avoid filesystem trashing with loose objects */
1222 if (!a
->in_pack
&& !b
->in_pack
)
1223 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1225 if (a
->in_pack
< b
->in_pack
)
1227 if (a
->in_pack
> b
->in_pack
)
1229 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1230 (a
->in_pack_offset
> b
->in_pack_offset
);
1233 static void get_object_details(void)
1236 struct object_entry
**sorted_by_offset
;
1238 sorted_by_offset
= xcalloc(nr_objects
, sizeof(struct object_entry
*));
1239 for (i
= 0; i
< nr_objects
; i
++)
1240 sorted_by_offset
[i
] = objects
+ i
;
1241 qsort(sorted_by_offset
, nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1244 for (i
= 0; i
< nr_objects
; i
++)
1245 check_object(sorted_by_offset
[i
]);
1246 free(sorted_by_offset
);
1250 * We search for deltas in a list sorted by type, by filename hash, and then
1251 * by size, so that we see progressively smaller and smaller files.
1252 * That's because we prefer deltas to be from the bigger file
1253 * to the smaller -- deletes are potentially cheaper, but perhaps
1254 * more importantly, the bigger file is likely the more recent
1255 * one. The deepest deltas are therefore the oldest objects which are
1256 * less susceptible to be accessed often.
1258 static int type_size_sort(const void *_a
, const void *_b
)
1260 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1261 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1263 if (a
->type
> b
->type
)
1265 if (a
->type
< b
->type
)
1267 if (a
->hash
> b
->hash
)
1269 if (a
->hash
< b
->hash
)
1271 if (a
->preferred_base
> b
->preferred_base
)
1273 if (a
->preferred_base
< b
->preferred_base
)
1275 if (a
->size
> b
->size
)
1277 if (a
->size
< b
->size
)
1279 return a
< b
? -1 : (a
> b
); /* newest first */
1283 struct object_entry
*entry
;
1285 struct delta_index
*index
;
1289 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1290 unsigned long delta_size
)
1292 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1295 if (delta_size
< cache_max_small_delta_size
)
1298 /* cache delta, if objects are large enough compared to delta size */
1299 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1305 #ifdef THREADED_DELTA_SEARCH
1307 static pthread_mutex_t read_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1308 #define read_lock() pthread_mutex_lock(&read_mutex)
1309 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1311 static pthread_mutex_t cache_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1312 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1313 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1315 static pthread_mutex_t progress_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1316 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1317 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1321 #define read_lock() (void)0
1322 #define read_unlock() (void)0
1323 #define cache_lock() (void)0
1324 #define cache_unlock() (void)0
1325 #define progress_lock() (void)0
1326 #define progress_unlock() (void)0
1330 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1331 unsigned max_depth
, unsigned long *mem_usage
)
1333 struct object_entry
*trg_entry
= trg
->entry
;
1334 struct object_entry
*src_entry
= src
->entry
;
1335 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1337 enum object_type type
;
1340 /* Don't bother doing diffs between different types */
1341 if (trg_entry
->type
!= src_entry
->type
)
1345 * We do not bother to try a delta that we discarded
1346 * on an earlier try, but only when reusing delta data.
1348 if (!no_reuse_delta
&& trg_entry
->in_pack
&&
1349 trg_entry
->in_pack
== src_entry
->in_pack
&&
1350 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1351 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1354 /* Let's not bust the allowed depth. */
1355 if (src
->depth
>= max_depth
)
1358 /* Now some size filtering heuristics. */
1359 trg_size
= trg_entry
->size
;
1360 if (!trg_entry
->delta
) {
1361 max_size
= trg_size
/2 - 20;
1364 max_size
= trg_entry
->delta_size
;
1365 ref_depth
= trg
->depth
;
1367 max_size
= max_size
* (max_depth
- src
->depth
) /
1368 (max_depth
- ref_depth
+ 1);
1371 src_size
= src_entry
->size
;
1372 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1373 if (sizediff
>= max_size
)
1375 if (trg_size
< src_size
/ 32)
1378 /* Load data if not already done */
1381 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1384 die("object %s cannot be read",
1385 sha1_to_hex(trg_entry
->idx
.sha1
));
1387 die("object %s inconsistent object length (%lu vs %lu)",
1388 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1393 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1396 die("object %s cannot be read",
1397 sha1_to_hex(src_entry
->idx
.sha1
));
1399 die("object %s inconsistent object length (%lu vs %lu)",
1400 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1404 src
->index
= create_delta_index(src
->data
, src_size
);
1406 static int warned
= 0;
1408 warning("suboptimal pack - out of memory");
1411 *mem_usage
+= sizeof_delta_index(src
->index
);
1414 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1418 if (trg_entry
->delta
) {
1419 /* Prefer only shallower same-sized deltas. */
1420 if (delta_size
== trg_entry
->delta_size
&&
1421 src
->depth
+ 1 >= trg
->depth
) {
1428 * Handle memory allocation outside of the cache
1429 * accounting lock. Compiler will optimize the strangeness
1430 * away when THREADED_DELTA_SEARCH is not defined.
1432 free(trg_entry
->delta_data
);
1434 if (trg_entry
->delta_data
) {
1435 delta_cache_size
-= trg_entry
->delta_size
;
1436 trg_entry
->delta_data
= NULL
;
1438 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1439 delta_cache_size
+= delta_size
;
1441 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1447 trg_entry
->delta
= src_entry
;
1448 trg_entry
->delta_size
= delta_size
;
1449 trg
->depth
= src
->depth
+ 1;
1454 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1456 struct object_entry
*child
= me
->delta_child
;
1459 unsigned int c
= check_delta_limit(child
, n
+ 1);
1462 child
= child
->delta_sibling
;
1467 static unsigned long free_unpacked(struct unpacked
*n
)
1469 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1470 free_delta_index(n
->index
);
1473 freed_mem
+= n
->entry
->size
;
1482 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1483 int window
, int depth
, unsigned *processed
)
1485 uint32_t i
, idx
= 0, count
= 0;
1486 unsigned int array_size
= window
* sizeof(struct unpacked
);
1487 struct unpacked
*array
;
1488 unsigned long mem_usage
= 0;
1490 array
= xmalloc(array_size
);
1491 memset(array
, 0, array_size
);
1494 struct object_entry
*entry
= *list
++;
1495 struct unpacked
*n
= array
+ idx
;
1496 int j
, max_depth
, best_base
= -1;
1504 if (!entry
->preferred_base
) {
1506 display_progress(progress_state
, *processed
);
1510 mem_usage
-= free_unpacked(n
);
1513 while (window_memory_limit
&&
1514 mem_usage
> window_memory_limit
&&
1516 uint32_t tail
= (idx
+ window
- count
) % window
;
1517 mem_usage
-= free_unpacked(array
+ tail
);
1521 /* We do not compute delta to *create* objects we are not
1524 if (entry
->preferred_base
)
1528 * If the current object is at pack edge, take the depth the
1529 * objects that depend on the current object into account
1530 * otherwise they would become too deep.
1533 if (entry
->delta_child
) {
1534 max_depth
-= check_delta_limit(entry
, 0);
1542 uint32_t other_idx
= idx
+ j
;
1544 if (other_idx
>= window
)
1545 other_idx
-= window
;
1546 m
= array
+ other_idx
;
1549 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1553 best_base
= other_idx
;
1556 /* if we made n a delta, and if n is already at max
1557 * depth, leaving it in the window is pointless. we
1558 * should evict it first.
1560 if (entry
->delta
&& depth
<= n
->depth
)
1564 * Move the best delta base up in the window, after the
1565 * currently deltified object, to keep it longer. It will
1566 * be the first base object to be attempted next.
1569 struct unpacked swap
= array
[best_base
];
1570 int dist
= (window
+ idx
- best_base
) % window
;
1571 int dst
= best_base
;
1573 int src
= (dst
+ 1) % window
;
1574 array
[dst
] = array
[src
];
1582 if (count
+ 1 < window
)
1588 for (i
= 0; i
< window
; ++i
) {
1589 free_delta_index(array
[i
].index
);
1590 free(array
[i
].data
);
1595 #ifdef THREADED_DELTA_SEARCH
1598 * The main thread waits on the condition that (at least) one of the workers
1599 * has stopped working (which is indicated in the .working member of
1600 * struct thread_params).
1601 * When a work thread has completed its work, it sets .working to 0 and
1602 * signals the main thread and waits on the condition that .data_ready
1606 struct thread_params
{
1608 struct object_entry
**list
;
1615 pthread_mutex_t mutex
;
1616 pthread_cond_t cond
;
1617 unsigned *processed
;
1620 static pthread_cond_t progress_cond
= PTHREAD_COND_INITIALIZER
;
1622 static void *threaded_find_deltas(void *arg
)
1624 struct thread_params
*me
= arg
;
1626 while (me
->remaining
) {
1627 find_deltas(me
->list
, &me
->remaining
,
1628 me
->window
, me
->depth
, me
->processed
);
1632 pthread_cond_signal(&progress_cond
);
1636 * We must not set ->data_ready before we wait on the
1637 * condition because the main thread may have set it to 1
1638 * before we get here. In order to be sure that new
1639 * work is available if we see 1 in ->data_ready, it
1640 * was initialized to 0 before this thread was spawned
1641 * and we reset it to 0 right away.
1643 pthread_mutex_lock(&me
->mutex
);
1644 while (!me
->data_ready
)
1645 pthread_cond_wait(&me
->cond
, &me
->mutex
);
1647 pthread_mutex_unlock(&me
->mutex
);
1649 /* leave ->working 1 so that this doesn't get more work assigned */
1653 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
1654 int window
, int depth
, unsigned *processed
)
1656 struct thread_params p
[delta_search_threads
];
1657 int i
, ret
, active_threads
= 0;
1659 if (delta_search_threads
<= 1) {
1660 find_deltas(list
, &list_size
, window
, depth
, processed
);
1664 /* Partition the work amongst work threads. */
1665 for (i
= 0; i
< delta_search_threads
; i
++) {
1666 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
1668 p
[i
].window
= window
;
1670 p
[i
].processed
= processed
;
1672 p
[i
].data_ready
= 0;
1674 /* try to split chunks on "path" boundaries */
1675 while (sub_size
&& sub_size
< list_size
&&
1676 list
[sub_size
]->hash
&&
1677 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
1681 p
[i
].list_size
= sub_size
;
1682 p
[i
].remaining
= sub_size
;
1685 list_size
-= sub_size
;
1688 /* Start work threads. */
1689 for (i
= 0; i
< delta_search_threads
; i
++) {
1690 if (!p
[i
].list_size
)
1692 pthread_mutex_init(&p
[i
].mutex
, NULL
);
1693 pthread_cond_init(&p
[i
].cond
, NULL
);
1694 ret
= pthread_create(&p
[i
].thread
, NULL
,
1695 threaded_find_deltas
, &p
[i
]);
1697 die("unable to create thread: %s", strerror(ret
));
1702 * Now let's wait for work completion. Each time a thread is done
1703 * with its work, we steal half of the remaining work from the
1704 * thread with the largest number of unprocessed objects and give
1705 * it to that newly idle thread. This ensure good load balancing
1706 * until the remaining object list segments are simply too short
1707 * to be worth splitting anymore.
1709 while (active_threads
) {
1710 struct thread_params
*target
= NULL
;
1711 struct thread_params
*victim
= NULL
;
1712 unsigned sub_size
= 0;
1716 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
1721 pthread_cond_wait(&progress_cond
, &progress_mutex
);
1724 for (i
= 0; i
< delta_search_threads
; i
++)
1725 if (p
[i
].remaining
> 2*window
&&
1726 (!victim
|| victim
->remaining
< p
[i
].remaining
))
1729 sub_size
= victim
->remaining
/ 2;
1730 list
= victim
->list
+ victim
->list_size
- sub_size
;
1731 while (sub_size
&& list
[0]->hash
&&
1732 list
[0]->hash
== list
[-1]->hash
) {
1738 * It is possible for some "paths" to have
1739 * so many objects that no hash boundary
1740 * might be found. Let's just steal the
1741 * exact half in that case.
1743 sub_size
= victim
->remaining
/ 2;
1746 target
->list
= list
;
1747 victim
->list_size
-= sub_size
;
1748 victim
->remaining
-= sub_size
;
1750 target
->list_size
= sub_size
;
1751 target
->remaining
= sub_size
;
1752 target
->working
= 1;
1755 pthread_mutex_lock(&target
->mutex
);
1756 target
->data_ready
= 1;
1757 pthread_cond_signal(&target
->cond
);
1758 pthread_mutex_unlock(&target
->mutex
);
1761 pthread_join(target
->thread
, NULL
);
1762 pthread_cond_destroy(&target
->cond
);
1763 pthread_mutex_destroy(&target
->mutex
);
1770 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1773 static void prepare_pack(int window
, int depth
)
1775 struct object_entry
**delta_list
;
1776 uint32_t i
, n
, nr_deltas
;
1778 get_object_details();
1780 if (!nr_objects
|| !window
|| !depth
)
1783 delta_list
= xmalloc(nr_objects
* sizeof(*delta_list
));
1786 for (i
= 0; i
< nr_objects
; i
++) {
1787 struct object_entry
*entry
= objects
+ i
;
1790 /* This happens if we decided to reuse existing
1791 * delta from a pack. "!no_reuse_delta &&" is implied.
1795 if (entry
->size
< 50)
1798 if (entry
->no_try_delta
)
1801 if (!entry
->preferred_base
)
1804 delta_list
[n
++] = entry
;
1807 if (nr_deltas
&& n
> 1) {
1808 unsigned nr_done
= 0;
1810 progress_state
= start_progress("Compressing objects",
1812 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
1813 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
1814 stop_progress(&progress_state
);
1815 if (nr_done
!= nr_deltas
)
1816 die("inconsistency with delta count");
1821 static int git_pack_config(const char *k
, const char *v
)
1823 if(!strcmp(k
, "pack.window")) {
1824 window
= git_config_int(k
, v
);
1827 if (!strcmp(k
, "pack.windowmemory")) {
1828 window_memory_limit
= git_config_ulong(k
, v
);
1831 if (!strcmp(k
, "pack.depth")) {
1832 depth
= git_config_int(k
, v
);
1835 if (!strcmp(k
, "pack.compression")) {
1836 int level
= git_config_int(k
, v
);
1838 level
= Z_DEFAULT_COMPRESSION
;
1839 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
1840 die("bad pack compression level %d", level
);
1841 pack_compression_level
= level
;
1842 pack_compression_seen
= 1;
1845 if (!strcmp(k
, "pack.deltacachesize")) {
1846 max_delta_cache_size
= git_config_int(k
, v
);
1849 if (!strcmp(k
, "pack.deltacachelimit")) {
1850 cache_max_small_delta_size
= git_config_int(k
, v
);
1853 if (!strcmp(k
, "pack.threads")) {
1854 delta_search_threads
= git_config_int(k
, v
);
1855 if (delta_search_threads
< 0)
1856 die("invalid number of threads specified (%d)",
1857 delta_search_threads
);
1858 #ifndef THREADED_DELTA_SEARCH
1859 if (delta_search_threads
!= 1)
1860 warning("no threads support, ignoring %s", k
);
1864 if (!strcmp(k
, "pack.indexversion")) {
1865 pack_idx_default_version
= git_config_int(k
, v
);
1866 if (pack_idx_default_version
> 2)
1867 die("bad pack.indexversion=%d", pack_idx_default_version
);
1870 if (!strcmp(k
, "pack.packsizelimit")) {
1871 pack_size_limit_cfg
= git_config_ulong(k
, v
);
1874 return git_default_config(k
, v
);
1877 static void read_object_list_from_stdin(void)
1879 char line
[40 + 1 + PATH_MAX
+ 2];
1880 unsigned char sha1
[20];
1883 if (!fgets(line
, sizeof(line
), stdin
)) {
1887 die("fgets returned NULL, not EOF, not error!");
1889 die("fgets: %s", strerror(errno
));
1893 if (line
[0] == '-') {
1894 if (get_sha1_hex(line
+1, sha1
))
1895 die("expected edge sha1, got garbage:\n %s",
1897 add_preferred_base(sha1
);
1900 if (get_sha1_hex(line
, sha1
))
1901 die("expected sha1, got garbage:\n %s", line
);
1903 add_preferred_base_object(line
+41);
1904 add_object_entry(sha1
, 0, line
+41, 0);
1908 #define OBJECT_ADDED (1u<<20)
1910 static void show_commit(struct commit
*commit
)
1912 add_object_entry(commit
->object
.sha1
, OBJ_COMMIT
, NULL
, 0);
1913 commit
->object
.flags
|= OBJECT_ADDED
;
1916 static void show_object(struct object_array_entry
*p
)
1918 add_preferred_base_object(p
->name
);
1919 add_object_entry(p
->item
->sha1
, p
->item
->type
, p
->name
, 0);
1920 p
->item
->flags
|= OBJECT_ADDED
;
1923 static void show_edge(struct commit
*commit
)
1925 add_preferred_base(commit
->object
.sha1
);
1928 struct in_pack_object
{
1930 struct object
*object
;
1936 struct in_pack_object
*array
;
1939 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
1941 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->sha1
, p
);
1942 in_pack
->array
[in_pack
->nr
].object
= object
;
1947 * Compare the objects in the offset order, in order to emulate the
1948 * "git-rev-list --objects" output that produced the pack originally.
1950 static int ofscmp(const void *a_
, const void *b_
)
1952 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
1953 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
1955 if (a
->offset
< b
->offset
)
1957 else if (a
->offset
> b
->offset
)
1960 return hashcmp(a
->object
->sha1
, b
->object
->sha1
);
1963 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
1965 struct packed_git
*p
;
1966 struct in_pack in_pack
;
1969 memset(&in_pack
, 0, sizeof(in_pack
));
1971 for (p
= packed_git
; p
; p
= p
->next
) {
1972 const unsigned char *sha1
;
1975 for (i
= 0; i
< revs
->num_ignore_packed
; i
++) {
1976 if (matches_pack_name(p
, revs
->ignore_packed
[i
]))
1979 if (revs
->num_ignore_packed
<= i
)
1981 if (open_pack_index(p
))
1982 die("cannot open pack index");
1984 ALLOC_GROW(in_pack
.array
,
1985 in_pack
.nr
+ p
->num_objects
,
1988 for (i
= 0; i
< p
->num_objects
; i
++) {
1989 sha1
= nth_packed_object_sha1(p
, i
);
1990 o
= lookup_unknown_object(sha1
);
1991 if (!(o
->flags
& OBJECT_ADDED
))
1992 mark_in_pack_object(o
, p
, &in_pack
);
1993 o
->flags
|= OBJECT_ADDED
;
1998 qsort(in_pack
.array
, in_pack
.nr
, sizeof(in_pack
.array
[0]),
2000 for (i
= 0; i
< in_pack
.nr
; i
++) {
2001 struct object
*o
= in_pack
.array
[i
].object
;
2002 add_object_entry(o
->sha1
, o
->type
, "", 0);
2005 free(in_pack
.array
);
2008 static void get_object_list(int ac
, const char **av
)
2010 struct rev_info revs
;
2014 init_revisions(&revs
, NULL
);
2015 save_commit_buffer
= 0;
2016 setup_revisions(ac
, av
, &revs
, NULL
);
2018 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2019 int len
= strlen(line
);
2020 if (len
&& line
[len
- 1] == '\n')
2025 if (!strcmp(line
, "--not")) {
2026 flags
^= UNINTERESTING
;
2029 die("not a rev '%s'", line
);
2031 if (handle_revision_arg(line
, &revs
, flags
, 1))
2032 die("bad revision '%s'", line
);
2035 if (prepare_revision_walk(&revs
))
2036 die("revision walk setup failed");
2037 mark_edges_uninteresting(revs
.commits
, &revs
, show_edge
);
2038 traverse_commit_list(&revs
, show_commit
, show_object
);
2040 if (keep_unreachable
)
2041 add_objects_in_unpacked_packs(&revs
);
2044 static int adjust_perm(const char *path
, mode_t mode
)
2046 if (chmod(path
, mode
))
2048 return adjust_shared_perm(path
);
2051 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
2053 int use_internal_rev_list
= 0;
2057 int rp_ac_alloc
= 64;
2060 rp_av
= xcalloc(rp_ac_alloc
, sizeof(*rp_av
));
2062 rp_av
[0] = "pack-objects";
2063 rp_av
[1] = "--objects"; /* --thin will make it --objects-edge */
2066 git_config(git_pack_config
);
2067 if (!pack_compression_seen
&& core_compression_seen
)
2068 pack_compression_level
= core_compression_level
;
2070 progress
= isatty(2);
2071 for (i
= 1; i
< argc
; i
++) {
2072 const char *arg
= argv
[i
];
2077 if (!strcmp("--non-empty", arg
)) {
2081 if (!strcmp("--local", arg
)) {
2085 if (!strcmp("--incremental", arg
)) {
2089 if (!prefixcmp(arg
, "--compression=")) {
2091 int level
= strtoul(arg
+14, &end
, 0);
2092 if (!arg
[14] || *end
)
2095 level
= Z_DEFAULT_COMPRESSION
;
2096 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
2097 die("bad pack compression level %d", level
);
2098 pack_compression_level
= level
;
2101 if (!prefixcmp(arg
, "--max-pack-size=")) {
2103 pack_size_limit_cfg
= 0;
2104 pack_size_limit
= strtoul(arg
+16, &end
, 0) * 1024 * 1024;
2105 if (!arg
[16] || *end
)
2109 if (!prefixcmp(arg
, "--window=")) {
2111 window
= strtoul(arg
+9, &end
, 0);
2112 if (!arg
[9] || *end
)
2116 if (!prefixcmp(arg
, "--window-memory=")) {
2117 if (!git_parse_ulong(arg
+16, &window_memory_limit
))
2121 if (!prefixcmp(arg
, "--threads=")) {
2123 delta_search_threads
= strtoul(arg
+10, &end
, 0);
2124 if (!arg
[10] || *end
|| delta_search_threads
< 0)
2126 #ifndef THREADED_DELTA_SEARCH
2127 if (delta_search_threads
!= 1)
2128 warning("no threads support, "
2129 "ignoring %s", arg
);
2133 if (!prefixcmp(arg
, "--depth=")) {
2135 depth
= strtoul(arg
+8, &end
, 0);
2136 if (!arg
[8] || *end
)
2140 if (!strcmp("--progress", arg
)) {
2144 if (!strcmp("--all-progress", arg
)) {
2148 if (!strcmp("-q", arg
)) {
2152 if (!strcmp("--no-reuse-delta", arg
)) {
2156 if (!strcmp("--no-reuse-object", arg
)) {
2157 no_reuse_object
= no_reuse_delta
= 1;
2160 if (!strcmp("--delta-base-offset", arg
)) {
2161 allow_ofs_delta
= 1;
2164 if (!strcmp("--stdout", arg
)) {
2168 if (!strcmp("--revs", arg
)) {
2169 use_internal_rev_list
= 1;
2172 if (!strcmp("--keep-unreachable", arg
)) {
2173 keep_unreachable
= 1;
2176 if (!strcmp("--unpacked", arg
) ||
2177 !prefixcmp(arg
, "--unpacked=") ||
2178 !strcmp("--reflog", arg
) ||
2179 !strcmp("--all", arg
)) {
2180 use_internal_rev_list
= 1;
2181 if (rp_ac
>= rp_ac_alloc
- 1) {
2182 rp_ac_alloc
= alloc_nr(rp_ac_alloc
);
2183 rp_av
= xrealloc(rp_av
,
2184 rp_ac_alloc
* sizeof(*rp_av
));
2186 rp_av
[rp_ac
++] = arg
;
2189 if (!strcmp("--thin", arg
)) {
2190 use_internal_rev_list
= 1;
2192 rp_av
[1] = "--objects-edge";
2195 if (!prefixcmp(arg
, "--index-version=")) {
2197 pack_idx_default_version
= strtoul(arg
+ 16, &c
, 10);
2198 if (pack_idx_default_version
> 2)
2201 pack_idx_off32_limit
= strtoul(c
+1, &c
, 0);
2202 if (*c
|| pack_idx_off32_limit
& 0x80000000)
2209 /* Traditionally "pack-objects [options] base extra" failed;
2210 * we would however want to take refs parameter that would
2211 * have been given to upstream rev-list ourselves, which means
2212 * we somehow want to say what the base name is. So the
2215 * pack-objects [options] base <refs...>
2217 * in other words, we would treat the first non-option as the
2218 * base_name and send everything else to the internal revision
2222 if (!pack_to_stdout
)
2223 base_name
= argv
[i
++];
2225 if (pack_to_stdout
!= !base_name
)
2228 if (!pack_to_stdout
&& !pack_size_limit
)
2229 pack_size_limit
= pack_size_limit_cfg
;
2231 if (pack_to_stdout
&& pack_size_limit
)
2232 die("--max-pack-size cannot be used to build a pack for transfer.");
2234 if (!pack_to_stdout
&& thin
)
2235 die("--thin cannot be used to build an indexable pack.");
2237 #ifdef THREADED_DELTA_SEARCH
2238 if (!delta_search_threads
) /* --threads=0 means autodetect */
2239 delta_search_threads
= online_cpus();
2242 prepare_packed_git();
2245 progress_state
= start_progress("Counting objects", 0);
2246 if (!use_internal_rev_list
)
2247 read_object_list_from_stdin();
2249 rp_av
[rp_ac
] = NULL
;
2250 get_object_list(rp_ac
, rp_av
);
2252 stop_progress(&progress_state
);
2254 if (non_empty
&& !nr_result
)
2257 prepare_pack(window
, depth
);
2260 fprintf(stderr
, "Total %u (delta %u), reused %u (delta %u)\n",
2261 written
, written_delta
, reused
, reused_delta
);