11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
16 #include "list-objects.h"
20 #ifdef THREADED_DELTA_SEARCH
21 #include "thread-utils.h"
25 static const char pack_usage
[] = "\
26 git pack-objects [{ -q | --progress | --all-progress }] \n\
27 [--max-pack-size=N] [--local] [--incremental] \n\
28 [--window=N] [--window-memory=N] [--depth=N] \n\
29 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
30 [--threads=N] [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
31 [--stdout | base-name] [--include-tag] \n\
32 [--keep-unreachable | --unpack-unreachable] \n\
33 [<ref-list | <object-list]";
36 struct pack_idx_entry idx
;
37 unsigned long size
; /* uncompressed size */
38 struct packed_git
*in_pack
; /* already in pack */
40 struct object_entry
*delta
; /* delta base object */
41 struct object_entry
*delta_child
; /* deltified objects who bases me */
42 struct object_entry
*delta_sibling
; /* other deltified objects who
43 * uses the same base as me
45 void *delta_data
; /* cached delta (uncompressed) */
46 unsigned long delta_size
; /* delta data size (uncompressed) */
47 unsigned long z_delta_size
; /* delta data size (compressed) */
48 unsigned int hash
; /* name hint hash */
49 enum object_type type
;
50 enum object_type in_pack_type
; /* could be delta */
51 unsigned char in_pack_header_size
;
52 unsigned char preferred_base
; /* we do not pack this, but is available
53 * to be used as the base object to delta
56 unsigned char no_try_delta
;
60 * Objects we are going to pack are collected in objects array (dynamically
61 * expanded). nr_objects & nr_alloc controls this array. They are stored
62 * in the order we see -- typically rev-list --objects order that gives us
63 * nice "minimum seek" order.
65 static struct object_entry
*objects
;
66 static struct pack_idx_entry
**written_list
;
67 static uint32_t nr_objects
, nr_alloc
, nr_result
, nr_written
;
70 static int reuse_delta
= 1, reuse_object
= 1;
71 static int keep_unreachable
, unpack_unreachable
, include_tag
;
73 static int incremental
;
74 static int allow_ofs_delta
;
75 static const char *base_name
;
76 static int progress
= 1;
77 static int window
= 10;
78 static uint32_t pack_size_limit
, pack_size_limit_cfg
;
79 static int depth
= 50;
80 static int delta_search_threads
= 1;
81 static int pack_to_stdout
;
82 static int num_preferred_base
;
83 static struct progress
*progress_state
;
84 static int pack_compression_level
= Z_DEFAULT_COMPRESSION
;
85 static int pack_compression_seen
;
87 static unsigned long delta_cache_size
= 0;
88 static unsigned long max_delta_cache_size
= 0;
89 static unsigned long cache_max_small_delta_size
= 1000;
91 static unsigned long window_memory_limit
= 0;
94 * The object names in objects array are hashed with this hashtable,
95 * to help looking up the entry by object name.
96 * This hashtable is built after all the objects are seen.
98 static int *object_ix
;
99 static int object_ix_hashsz
;
104 static uint32_t written
, written_delta
;
105 static uint32_t reused
, reused_delta
;
108 static void *get_delta(struct object_entry
*entry
)
110 unsigned long size
, base_size
, delta_size
;
111 void *buf
, *base_buf
, *delta_buf
;
112 enum object_type type
;
114 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
116 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
117 base_buf
= read_sha1_file(entry
->delta
->idx
.sha1
, &type
, &base_size
);
119 die("unable to read %s", sha1_to_hex(entry
->delta
->idx
.sha1
));
120 delta_buf
= diff_delta(base_buf
, base_size
,
121 buf
, size
, &delta_size
, 0);
122 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
123 die("delta size changed");
129 static unsigned long do_compress(void **pptr
, unsigned long size
)
133 unsigned long maxsize
;
135 memset(&stream
, 0, sizeof(stream
));
136 deflateInit(&stream
, pack_compression_level
);
137 maxsize
= deflateBound(&stream
, size
);
140 out
= xmalloc(maxsize
);
144 stream
.avail_in
= size
;
145 stream
.next_out
= out
;
146 stream
.avail_out
= maxsize
;
147 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
152 return stream
.total_out
;
156 * The per-object header is a pretty dense thing, which is
157 * - first byte: low four bits are "size", then three bits of "type",
158 * and the high bit is "size continues".
159 * - each byte afterwards: low seven bits are size continuation,
160 * with the high bit being "size continues"
162 static int encode_header(enum object_type type
, unsigned long size
, unsigned char *hdr
)
167 if (type
< OBJ_COMMIT
|| type
> OBJ_REF_DELTA
)
168 die("bad type %d", type
);
170 c
= (type
<< 4) | (size
& 15);
183 * we are going to reuse the existing object data as is. make
184 * sure it is not corrupt.
186 static int check_pack_inflate(struct packed_git
*p
,
187 struct pack_window
**w_curs
,
190 unsigned long expect
)
193 unsigned char fakebuf
[4096], *in
;
196 memset(&stream
, 0, sizeof(stream
));
197 inflateInit(&stream
);
199 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
201 stream
.next_out
= fakebuf
;
202 stream
.avail_out
= sizeof(fakebuf
);
203 st
= inflate(&stream
, Z_FINISH
);
204 offset
+= stream
.next_in
- in
;
205 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
207 return (st
== Z_STREAM_END
&&
208 stream
.total_out
== expect
&&
209 stream
.total_in
== len
) ? 0 : -1;
212 static void copy_pack_data(struct sha1file
*f
,
213 struct packed_git
*p
,
214 struct pack_window
**w_curs
,
222 in
= use_pack(p
, w_curs
, offset
, &avail
);
224 avail
= (unsigned int)len
;
225 sha1write(f
, in
, avail
);
231 static unsigned long write_object(struct sha1file
*f
,
232 struct object_entry
*entry
,
235 unsigned long size
, limit
, datalen
;
237 unsigned char header
[10], dheader
[10];
239 enum object_type type
;
240 int usable_delta
, to_reuse
;
247 /* write limit if limited packsize and not first object */
248 limit
= pack_size_limit
&& nr_written
?
249 pack_size_limit
- write_offset
: 0;
252 usable_delta
= 0; /* no delta */
253 else if (!pack_size_limit
)
254 usable_delta
= 1; /* unlimited packfile */
255 else if (entry
->delta
->idx
.offset
== (off_t
)-1)
256 usable_delta
= 0; /* base was written to another pack */
257 else if (entry
->delta
->idx
.offset
)
258 usable_delta
= 1; /* base already exists in this pack */
260 usable_delta
= 0; /* base could end up in another pack */
263 to_reuse
= 0; /* explicit */
264 else if (!entry
->in_pack
)
265 to_reuse
= 0; /* can't reuse what we don't have */
266 else if (type
== OBJ_REF_DELTA
|| type
== OBJ_OFS_DELTA
)
267 /* check_object() decided it for us ... */
268 to_reuse
= usable_delta
;
269 /* ... but pack split may override that */
270 else if (type
!= entry
->in_pack_type
)
271 to_reuse
= 0; /* pack has delta which is unusable */
272 else if (entry
->delta
)
273 to_reuse
= 0; /* we want to pack afresh */
275 to_reuse
= 1; /* we have it in-pack undeltified,
276 * and we do not need to deltify it.
282 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
284 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
286 * make sure no cached delta data remains from a
287 * previous attempt before a pack split occured.
289 free(entry
->delta_data
);
290 entry
->delta_data
= NULL
;
291 entry
->z_delta_size
= 0;
292 } else if (entry
->delta_data
) {
293 size
= entry
->delta_size
;
294 buf
= entry
->delta_data
;
295 entry
->delta_data
= NULL
;
296 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
297 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
299 buf
= get_delta(entry
);
300 size
= entry
->delta_size
;
301 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
302 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
305 if (entry
->z_delta_size
)
306 datalen
= entry
->z_delta_size
;
308 datalen
= do_compress(&buf
, size
);
311 * The object header is a byte of 'type' followed by zero or
312 * more bytes of length.
314 hdrlen
= encode_header(type
, size
, header
);
316 if (type
== OBJ_OFS_DELTA
) {
318 * Deltas with relative base contain an additional
319 * encoding of the relative offset for the delta
320 * base from this object's position in the pack.
322 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
323 unsigned pos
= sizeof(dheader
) - 1;
324 dheader
[pos
] = ofs
& 127;
326 dheader
[--pos
] = 128 | (--ofs
& 127);
327 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
331 sha1write(f
, header
, hdrlen
);
332 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
333 hdrlen
+= sizeof(dheader
) - pos
;
334 } else if (type
== OBJ_REF_DELTA
) {
336 * Deltas with a base reference contain
337 * an additional 20 bytes for the base sha1.
339 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
343 sha1write(f
, header
, hdrlen
);
344 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
347 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
351 sha1write(f
, header
, hdrlen
);
353 sha1write(f
, buf
, datalen
);
357 struct packed_git
*p
= entry
->in_pack
;
358 struct pack_window
*w_curs
= NULL
;
359 struct revindex_entry
*revidx
;
363 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
364 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
365 hdrlen
= encode_header(type
, entry
->size
, header
);
367 offset
= entry
->in_pack_offset
;
368 revidx
= find_pack_revindex(p
, offset
);
369 datalen
= revidx
[1].offset
- offset
;
370 if (!pack_to_stdout
&& p
->index_version
> 1 &&
371 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
)) {
372 error("bad packed object CRC for %s", sha1_to_hex(entry
->idx
.sha1
));
377 offset
+= entry
->in_pack_header_size
;
378 datalen
-= entry
->in_pack_header_size
;
379 if (!pack_to_stdout
&& p
->index_version
== 1 &&
380 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry
->size
)) {
381 error("corrupt packed object for %s", sha1_to_hex(entry
->idx
.sha1
));
386 if (type
== OBJ_OFS_DELTA
) {
387 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
388 unsigned pos
= sizeof(dheader
) - 1;
389 dheader
[pos
] = ofs
& 127;
391 dheader
[--pos
] = 128 | (--ofs
& 127);
392 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
396 sha1write(f
, header
, hdrlen
);
397 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
398 hdrlen
+= sizeof(dheader
) - pos
;
400 } else if (type
== OBJ_REF_DELTA
) {
401 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
405 sha1write(f
, header
, hdrlen
);
406 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
410 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
414 sha1write(f
, header
, hdrlen
);
416 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
424 entry
->idx
.crc32
= crc32_end(f
);
425 return hdrlen
+ datalen
;
428 static int write_one(struct sha1file
*f
,
429 struct object_entry
*e
,
434 /* offset is non zero if object is written already. */
435 if (e
->idx
.offset
|| e
->preferred_base
)
438 /* if we are deltified, write out base object first. */
439 if (e
->delta
&& !write_one(f
, e
->delta
, offset
))
442 e
->idx
.offset
= *offset
;
443 size
= write_object(f
, e
, *offset
);
448 written_list
[nr_written
++] = &e
->idx
;
450 /* make sure off_t is sufficiently large not to wrap */
451 if (*offset
> *offset
+ size
)
452 die("pack too large for current definition of off_t");
457 /* forward declaration for write_pack_file */
458 static int adjust_perm(const char *path
, mode_t mode
);
460 static void write_pack_file(void)
465 struct pack_header hdr
;
466 uint32_t nr_remaining
= nr_result
;
467 time_t last_mtime
= 0;
469 if (progress
> pack_to_stdout
)
470 progress_state
= start_progress("Writing objects", nr_result
);
471 written_list
= xmalloc(nr_objects
* sizeof(*written_list
));
474 unsigned char sha1
[20];
475 char *pack_tmp_name
= NULL
;
477 if (pack_to_stdout
) {
478 f
= sha1fd_throughput(1, "<stdout>", progress_state
);
480 char tmpname
[PATH_MAX
];
482 snprintf(tmpname
, sizeof(tmpname
),
483 "%s/pack/tmp_pack_XXXXXX", get_object_directory());
484 fd
= xmkstemp(tmpname
);
485 pack_tmp_name
= xstrdup(tmpname
);
486 f
= sha1fd(fd
, pack_tmp_name
);
489 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
490 hdr
.hdr_version
= htonl(PACK_VERSION
);
491 hdr
.hdr_entries
= htonl(nr_remaining
);
492 sha1write(f
, &hdr
, sizeof(hdr
));
493 offset
= sizeof(hdr
);
495 for (; i
< nr_objects
; i
++) {
496 if (!write_one(f
, objects
+ i
, &offset
))
498 display_progress(progress_state
, written
);
502 * Did we write the wrong # entries in the header?
503 * If so, rewrite it like in fast-import
505 if (pack_to_stdout
) {
506 sha1close(f
, sha1
, CSUM_CLOSE
);
507 } else if (nr_written
== nr_remaining
) {
508 sha1close(f
, sha1
, CSUM_FSYNC
);
510 int fd
= sha1close(f
, sha1
, 0);
511 fixup_pack_header_footer(fd
, sha1
, pack_tmp_name
,
512 nr_written
, sha1
, offset
);
516 if (!pack_to_stdout
) {
517 mode_t mode
= umask(0);
519 char *idx_tmp_name
, tmpname
[PATH_MAX
];
524 idx_tmp_name
= write_idx_file(NULL
, written_list
,
527 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.pack",
528 base_name
, sha1_to_hex(sha1
));
529 if (adjust_perm(pack_tmp_name
, mode
))
530 die("unable to make temporary pack file readable: %s",
532 if (rename(pack_tmp_name
, tmpname
))
533 die("unable to rename temporary pack file: %s",
537 * Packs are runtime accessed in their mtime
538 * order since newer packs are more likely to contain
539 * younger objects. So if we are creating multiple
540 * packs then we should modify the mtime of later ones
541 * to preserve this property.
543 if (stat(tmpname
, &st
) < 0) {
544 warning("failed to stat %s: %s",
545 tmpname
, strerror(errno
));
546 } else if (!last_mtime
) {
547 last_mtime
= st
.st_mtime
;
550 utb
.actime
= st
.st_atime
;
551 utb
.modtime
= --last_mtime
;
552 if (utime(tmpname
, &utb
) < 0)
553 warning("failed utime() on %s: %s",
554 tmpname
, strerror(errno
));
557 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.idx",
558 base_name
, sha1_to_hex(sha1
));
559 if (adjust_perm(idx_tmp_name
, mode
))
560 die("unable to make temporary index file readable: %s",
562 if (rename(idx_tmp_name
, tmpname
))
563 die("unable to rename temporary index file: %s",
568 puts(sha1_to_hex(sha1
));
571 /* mark written objects as written to previous pack */
572 for (j
= 0; j
< nr_written
; j
++) {
573 written_list
[j
]->offset
= (off_t
)-1;
575 nr_remaining
-= nr_written
;
576 } while (nr_remaining
&& i
< nr_objects
);
579 stop_progress(&progress_state
);
580 if (written
!= nr_result
)
581 die("wrote %"PRIu32
" objects while expecting %"PRIu32
,
584 * We have scanned through [0 ... i). Since we have written
585 * the correct number of objects, the remaining [i ... nr_objects)
586 * items must be either already written (due to out-of-order delta base)
587 * or a preferred base. Count those which are neither and complain if any.
589 for (j
= 0; i
< nr_objects
; i
++) {
590 struct object_entry
*e
= objects
+ i
;
591 j
+= !e
->idx
.offset
&& !e
->preferred_base
;
594 die("wrote %"PRIu32
" objects as expected but %"PRIu32
595 " unwritten", written
, j
);
598 static int locate_object_entry_hash(const unsigned char *sha1
)
602 memcpy(&ui
, sha1
, sizeof(unsigned int));
603 i
= ui
% object_ix_hashsz
;
604 while (0 < object_ix
[i
]) {
605 if (!hashcmp(sha1
, objects
[object_ix
[i
] - 1].idx
.sha1
))
607 if (++i
== object_ix_hashsz
)
613 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
617 if (!object_ix_hashsz
)
620 i
= locate_object_entry_hash(sha1
);
622 return &objects
[object_ix
[i
]-1];
626 static void rehash_objects(void)
629 struct object_entry
*oe
;
631 object_ix_hashsz
= nr_objects
* 3;
632 if (object_ix_hashsz
< 1024)
633 object_ix_hashsz
= 1024;
634 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
635 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
636 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
637 int ix
= locate_object_entry_hash(oe
->idx
.sha1
);
641 object_ix
[ix
] = i
+ 1;
645 static unsigned name_hash(const char *name
)
654 * This effectively just creates a sortable number from the
655 * last sixteen non-whitespace characters. Last characters
656 * count "most", so things that end in ".c" sort together.
658 while ((c
= *name
++) != 0) {
661 hash
= (hash
>> 2) + (c
<< 24);
666 static void setup_delta_attr_check(struct git_attr_check
*check
)
668 static struct git_attr
*attr_delta
;
671 attr_delta
= git_attr("delta", 5);
673 check
[0].attr
= attr_delta
;
676 static int no_try_delta(const char *path
)
678 struct git_attr_check check
[1];
680 setup_delta_attr_check(check
);
681 if (git_checkattr(path
, ARRAY_SIZE(check
), check
))
683 if (ATTR_FALSE(check
->value
))
688 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
689 const char *name
, int exclude
)
691 struct object_entry
*entry
;
692 struct packed_git
*p
, *found_pack
= NULL
;
693 off_t found_offset
= 0;
695 unsigned hash
= name_hash(name
);
697 ix
= nr_objects
? locate_object_entry_hash(sha1
) : -1;
700 entry
= objects
+ object_ix
[ix
] - 1;
701 if (!entry
->preferred_base
)
703 entry
->preferred_base
= 1;
708 for (p
= packed_git
; p
; p
= p
->next
) {
709 off_t offset
= find_pack_entry_one(sha1
, p
);
712 found_offset
= offset
;
719 if (local
&& !p
->pack_local
)
724 if (nr_objects
>= nr_alloc
) {
725 nr_alloc
= (nr_alloc
+ 1024) * 3 / 2;
726 objects
= xrealloc(objects
, nr_alloc
* sizeof(*entry
));
729 entry
= objects
+ nr_objects
++;
730 memset(entry
, 0, sizeof(*entry
));
731 hashcpy(entry
->idx
.sha1
, sha1
);
736 entry
->preferred_base
= 1;
740 entry
->in_pack
= found_pack
;
741 entry
->in_pack_offset
= found_offset
;
744 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
747 object_ix
[-1 - ix
] = nr_objects
;
749 display_progress(progress_state
, nr_objects
);
751 if (name
&& no_try_delta(name
))
752 entry
->no_try_delta
= 1;
757 struct pbase_tree_cache
{
758 unsigned char sha1
[20];
762 unsigned long tree_size
;
765 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
766 static int pbase_tree_cache_ix(const unsigned char *sha1
)
768 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
770 static int pbase_tree_cache_ix_incr(int ix
)
772 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
775 static struct pbase_tree
{
776 struct pbase_tree
*next
;
777 /* This is a phony "cache" entry; we are not
778 * going to evict it nor find it through _get()
779 * mechanism -- this is for the toplevel node that
780 * would almost always change with any commit.
782 struct pbase_tree_cache pcache
;
785 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
787 struct pbase_tree_cache
*ent
, *nent
;
790 enum object_type type
;
792 int my_ix
= pbase_tree_cache_ix(sha1
);
793 int available_ix
= -1;
795 /* pbase-tree-cache acts as a limited hashtable.
796 * your object will be found at your index or within a few
797 * slots after that slot if it is cached.
799 for (neigh
= 0; neigh
< 8; neigh
++) {
800 ent
= pbase_tree_cache
[my_ix
];
801 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
805 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
806 ((0 <= available_ix
) &&
807 (!ent
&& pbase_tree_cache
[available_ix
])))
808 available_ix
= my_ix
;
811 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
814 /* Did not find one. Either we got a bogus request or
815 * we need to read and perhaps cache.
817 data
= read_sha1_file(sha1
, &type
, &size
);
820 if (type
!= OBJ_TREE
) {
825 /* We need to either cache or return a throwaway copy */
827 if (available_ix
< 0)
830 ent
= pbase_tree_cache
[available_ix
];
831 my_ix
= available_ix
;
835 nent
= xmalloc(sizeof(*nent
));
836 nent
->temporary
= (available_ix
< 0);
839 /* evict and reuse */
840 free(ent
->tree_data
);
843 hashcpy(nent
->sha1
, sha1
);
844 nent
->tree_data
= data
;
845 nent
->tree_size
= size
;
847 if (!nent
->temporary
)
848 pbase_tree_cache
[my_ix
] = nent
;
852 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
854 if (!cache
->temporary
) {
858 free(cache
->tree_data
);
862 static int name_cmp_len(const char *name
)
865 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
870 static void add_pbase_object(struct tree_desc
*tree
,
873 const char *fullname
)
875 struct name_entry entry
;
878 while (tree_entry(tree
,&entry
)) {
879 if (S_ISGITLINK(entry
.mode
))
881 cmp
= tree_entry_len(entry
.path
, entry
.sha1
) != cmplen
? 1 :
882 memcmp(name
, entry
.path
, cmplen
);
887 if (name
[cmplen
] != '/') {
888 add_object_entry(entry
.sha1
,
889 object_type(entry
.mode
),
893 if (S_ISDIR(entry
.mode
)) {
894 struct tree_desc sub
;
895 struct pbase_tree_cache
*tree
;
896 const char *down
= name
+cmplen
+1;
897 int downlen
= name_cmp_len(down
);
899 tree
= pbase_tree_get(entry
.sha1
);
902 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
904 add_pbase_object(&sub
, down
, downlen
, fullname
);
905 pbase_tree_put(tree
);
910 static unsigned *done_pbase_paths
;
911 static int done_pbase_paths_num
;
912 static int done_pbase_paths_alloc
;
913 static int done_pbase_path_pos(unsigned hash
)
916 int hi
= done_pbase_paths_num
;
918 int mi
= (hi
+ lo
) / 2;
919 if (done_pbase_paths
[mi
] == hash
)
921 if (done_pbase_paths
[mi
] < hash
)
929 static int check_pbase_path(unsigned hash
)
931 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
935 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
936 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
937 done_pbase_paths
= xrealloc(done_pbase_paths
,
938 done_pbase_paths_alloc
*
941 done_pbase_paths_num
++;
942 if (pos
< done_pbase_paths_num
)
943 memmove(done_pbase_paths
+ pos
+ 1,
944 done_pbase_paths
+ pos
,
945 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
946 done_pbase_paths
[pos
] = hash
;
950 static void add_preferred_base_object(const char *name
)
952 struct pbase_tree
*it
;
954 unsigned hash
= name_hash(name
);
956 if (!num_preferred_base
|| check_pbase_path(hash
))
959 cmplen
= name_cmp_len(name
);
960 for (it
= pbase_tree
; it
; it
= it
->next
) {
962 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
965 struct tree_desc tree
;
966 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
967 add_pbase_object(&tree
, name
, cmplen
, name
);
972 static void add_preferred_base(unsigned char *sha1
)
974 struct pbase_tree
*it
;
977 unsigned char tree_sha1
[20];
979 if (window
<= num_preferred_base
++)
982 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
986 for (it
= pbase_tree
; it
; it
= it
->next
) {
987 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
993 it
= xcalloc(1, sizeof(*it
));
994 it
->next
= pbase_tree
;
997 hashcpy(it
->pcache
.sha1
, tree_sha1
);
998 it
->pcache
.tree_data
= data
;
999 it
->pcache
.tree_size
= size
;
1002 static void check_object(struct object_entry
*entry
)
1004 if (entry
->in_pack
) {
1005 struct packed_git
*p
= entry
->in_pack
;
1006 struct pack_window
*w_curs
= NULL
;
1007 const unsigned char *base_ref
= NULL
;
1008 struct object_entry
*base_entry
;
1009 unsigned long used
, used_0
;
1012 unsigned char *buf
, c
;
1014 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1017 * We want in_pack_type even if we do not reuse delta
1018 * since non-delta representations could still be reused.
1020 used
= unpack_object_header_buffer(buf
, avail
,
1021 &entry
->in_pack_type
,
1027 * Determine if this is a delta and if so whether we can
1028 * reuse it or not. Otherwise let's find out as cheaply as
1029 * possible what the actual type and size for this object is.
1031 switch (entry
->in_pack_type
) {
1033 /* Not a delta hence we've already got all we need. */
1034 entry
->type
= entry
->in_pack_type
;
1035 entry
->in_pack_header_size
= used
;
1036 if (entry
->type
< OBJ_COMMIT
|| entry
->type
> OBJ_BLOB
)
1038 unuse_pack(&w_curs
);
1041 if (reuse_delta
&& !entry
->preferred_base
)
1042 base_ref
= use_pack(p
, &w_curs
,
1043 entry
->in_pack_offset
+ used
, NULL
);
1044 entry
->in_pack_header_size
= used
+ 20;
1047 buf
= use_pack(p
, &w_curs
,
1048 entry
->in_pack_offset
+ used
, NULL
);
1054 if (!ofs
|| MSB(ofs
, 7)) {
1055 error("delta base offset overflow in pack for %s",
1056 sha1_to_hex(entry
->idx
.sha1
));
1060 ofs
= (ofs
<< 7) + (c
& 127);
1062 ofs
= entry
->in_pack_offset
- ofs
;
1063 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1064 error("delta base offset out of bound for %s",
1065 sha1_to_hex(entry
->idx
.sha1
));
1068 if (reuse_delta
&& !entry
->preferred_base
) {
1069 struct revindex_entry
*revidx
;
1070 revidx
= find_pack_revindex(p
, ofs
);
1073 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1075 entry
->in_pack_header_size
= used
+ used_0
;
1079 if (base_ref
&& (base_entry
= locate_object_entry(base_ref
))) {
1081 * If base_ref was set above that means we wish to
1082 * reuse delta data, and we even found that base
1083 * in the list of objects we want to pack. Goodie!
1085 * Depth value does not matter - find_deltas() will
1086 * never consider reused delta as the base object to
1087 * deltify other objects against, in order to avoid
1090 entry
->type
= entry
->in_pack_type
;
1091 entry
->delta
= base_entry
;
1092 entry
->delta_size
= entry
->size
;
1093 entry
->delta_sibling
= base_entry
->delta_child
;
1094 base_entry
->delta_child
= entry
;
1095 unuse_pack(&w_curs
);
1101 * This must be a delta and we already know what the
1102 * final object type is. Let's extract the actual
1103 * object size from the delta header.
1105 entry
->size
= get_size_from_delta(p
, &w_curs
,
1106 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1107 if (entry
->size
== 0)
1109 unuse_pack(&w_curs
);
1114 * No choice but to fall back to the recursive delta walk
1115 * with sha1_object_info() to find about the object type
1119 unuse_pack(&w_curs
);
1122 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1124 * The error condition is checked in prepare_pack(). This is
1125 * to permit a missing preferred base object to be ignored
1126 * as a preferred base. Doing so can result in a larger
1127 * pack file, but the transfer will still take place.
1131 static int pack_offset_sort(const void *_a
, const void *_b
)
1133 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1134 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1136 /* avoid filesystem trashing with loose objects */
1137 if (!a
->in_pack
&& !b
->in_pack
)
1138 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1140 if (a
->in_pack
< b
->in_pack
)
1142 if (a
->in_pack
> b
->in_pack
)
1144 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1145 (a
->in_pack_offset
> b
->in_pack_offset
);
1148 static void get_object_details(void)
1151 struct object_entry
**sorted_by_offset
;
1153 sorted_by_offset
= xcalloc(nr_objects
, sizeof(struct object_entry
*));
1154 for (i
= 0; i
< nr_objects
; i
++)
1155 sorted_by_offset
[i
] = objects
+ i
;
1156 qsort(sorted_by_offset
, nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1158 for (i
= 0; i
< nr_objects
; i
++)
1159 check_object(sorted_by_offset
[i
]);
1161 free(sorted_by_offset
);
1165 * We search for deltas in a list sorted by type, by filename hash, and then
1166 * by size, so that we see progressively smaller and smaller files.
1167 * That's because we prefer deltas to be from the bigger file
1168 * to the smaller -- deletes are potentially cheaper, but perhaps
1169 * more importantly, the bigger file is likely the more recent
1170 * one. The deepest deltas are therefore the oldest objects which are
1171 * less susceptible to be accessed often.
1173 static int type_size_sort(const void *_a
, const void *_b
)
1175 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1176 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1178 if (a
->type
> b
->type
)
1180 if (a
->type
< b
->type
)
1182 if (a
->hash
> b
->hash
)
1184 if (a
->hash
< b
->hash
)
1186 if (a
->preferred_base
> b
->preferred_base
)
1188 if (a
->preferred_base
< b
->preferred_base
)
1190 if (a
->size
> b
->size
)
1192 if (a
->size
< b
->size
)
1194 return a
< b
? -1 : (a
> b
); /* newest first */
1198 struct object_entry
*entry
;
1200 struct delta_index
*index
;
1204 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1205 unsigned long delta_size
)
1207 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1210 if (delta_size
< cache_max_small_delta_size
)
1213 /* cache delta, if objects are large enough compared to delta size */
1214 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1220 #ifdef THREADED_DELTA_SEARCH
1222 static pthread_mutex_t read_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1223 #define read_lock() pthread_mutex_lock(&read_mutex)
1224 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1226 static pthread_mutex_t cache_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1227 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1228 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1230 static pthread_mutex_t progress_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1231 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1232 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1236 #define read_lock() (void)0
1237 #define read_unlock() (void)0
1238 #define cache_lock() (void)0
1239 #define cache_unlock() (void)0
1240 #define progress_lock() (void)0
1241 #define progress_unlock() (void)0
1245 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1246 unsigned max_depth
, unsigned long *mem_usage
)
1248 struct object_entry
*trg_entry
= trg
->entry
;
1249 struct object_entry
*src_entry
= src
->entry
;
1250 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1252 enum object_type type
;
1255 /* Don't bother doing diffs between different types */
1256 if (trg_entry
->type
!= src_entry
->type
)
1260 * We do not bother to try a delta that we discarded
1261 * on an earlier try, but only when reusing delta data.
1263 if (reuse_delta
&& trg_entry
->in_pack
&&
1264 trg_entry
->in_pack
== src_entry
->in_pack
&&
1265 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1266 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1269 /* Let's not bust the allowed depth. */
1270 if (src
->depth
>= max_depth
)
1273 /* Now some size filtering heuristics. */
1274 trg_size
= trg_entry
->size
;
1275 if (!trg_entry
->delta
) {
1276 max_size
= trg_size
/2 - 20;
1279 max_size
= trg_entry
->delta_size
;
1280 ref_depth
= trg
->depth
;
1282 max_size
= max_size
* (max_depth
- src
->depth
) /
1283 (max_depth
- ref_depth
+ 1);
1286 src_size
= src_entry
->size
;
1287 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1288 if (sizediff
>= max_size
)
1290 if (trg_size
< src_size
/ 32)
1293 /* Load data if not already done */
1296 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1299 die("object %s cannot be read",
1300 sha1_to_hex(trg_entry
->idx
.sha1
));
1302 die("object %s inconsistent object length (%lu vs %lu)",
1303 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1308 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1311 die("object %s cannot be read",
1312 sha1_to_hex(src_entry
->idx
.sha1
));
1314 die("object %s inconsistent object length (%lu vs %lu)",
1315 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1319 src
->index
= create_delta_index(src
->data
, src_size
);
1321 static int warned
= 0;
1323 warning("suboptimal pack - out of memory");
1326 *mem_usage
+= sizeof_delta_index(src
->index
);
1329 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1333 if (trg_entry
->delta
) {
1334 /* Prefer only shallower same-sized deltas. */
1335 if (delta_size
== trg_entry
->delta_size
&&
1336 src
->depth
+ 1 >= trg
->depth
) {
1343 * Handle memory allocation outside of the cache
1344 * accounting lock. Compiler will optimize the strangeness
1345 * away when THREADED_DELTA_SEARCH is not defined.
1347 free(trg_entry
->delta_data
);
1349 if (trg_entry
->delta_data
) {
1350 delta_cache_size
-= trg_entry
->delta_size
;
1351 trg_entry
->delta_data
= NULL
;
1353 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1354 delta_cache_size
+= delta_size
;
1356 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1362 trg_entry
->delta
= src_entry
;
1363 trg_entry
->delta_size
= delta_size
;
1364 trg
->depth
= src
->depth
+ 1;
1369 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1371 struct object_entry
*child
= me
->delta_child
;
1374 unsigned int c
= check_delta_limit(child
, n
+ 1);
1377 child
= child
->delta_sibling
;
1382 static unsigned long free_unpacked(struct unpacked
*n
)
1384 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1385 free_delta_index(n
->index
);
1388 freed_mem
+= n
->entry
->size
;
1397 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1398 int window
, int depth
, unsigned *processed
)
1400 uint32_t i
, idx
= 0, count
= 0;
1401 struct unpacked
*array
;
1402 unsigned long mem_usage
= 0;
1404 array
= xcalloc(window
, sizeof(struct unpacked
));
1407 struct object_entry
*entry
;
1408 struct unpacked
*n
= array
+ idx
;
1409 int j
, max_depth
, best_base
= -1;
1418 if (!entry
->preferred_base
) {
1420 display_progress(progress_state
, *processed
);
1424 mem_usage
-= free_unpacked(n
);
1427 while (window_memory_limit
&&
1428 mem_usage
> window_memory_limit
&&
1430 uint32_t tail
= (idx
+ window
- count
) % window
;
1431 mem_usage
-= free_unpacked(array
+ tail
);
1435 /* We do not compute delta to *create* objects we are not
1438 if (entry
->preferred_base
)
1442 * If the current object is at pack edge, take the depth the
1443 * objects that depend on the current object into account
1444 * otherwise they would become too deep.
1447 if (entry
->delta_child
) {
1448 max_depth
-= check_delta_limit(entry
, 0);
1456 uint32_t other_idx
= idx
+ j
;
1458 if (other_idx
>= window
)
1459 other_idx
-= window
;
1460 m
= array
+ other_idx
;
1463 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1467 best_base
= other_idx
;
1471 * If we decided to cache the delta data, then it is best
1472 * to compress it right away. First because we have to do
1473 * it anyway, and doing it here while we're threaded will
1474 * save a lot of time in the non threaded write phase,
1475 * as well as allow for caching more deltas within
1476 * the same cache size limit.
1478 * But only if not writing to stdout, since in that case
1479 * the network is most likely throttling writes anyway,
1480 * and therefore it is best to go to the write phase ASAP
1481 * instead, as we can afford spending more time compressing
1482 * between writes at that moment.
1484 if (entry
->delta_data
&& !pack_to_stdout
) {
1485 entry
->z_delta_size
= do_compress(&entry
->delta_data
,
1488 delta_cache_size
-= entry
->delta_size
;
1489 delta_cache_size
+= entry
->z_delta_size
;
1493 /* if we made n a delta, and if n is already at max
1494 * depth, leaving it in the window is pointless. we
1495 * should evict it first.
1497 if (entry
->delta
&& max_depth
<= n
->depth
)
1501 * Move the best delta base up in the window, after the
1502 * currently deltified object, to keep it longer. It will
1503 * be the first base object to be attempted next.
1506 struct unpacked swap
= array
[best_base
];
1507 int dist
= (window
+ idx
- best_base
) % window
;
1508 int dst
= best_base
;
1510 int src
= (dst
+ 1) % window
;
1511 array
[dst
] = array
[src
];
1519 if (count
+ 1 < window
)
1525 for (i
= 0; i
< window
; ++i
) {
1526 free_delta_index(array
[i
].index
);
1527 free(array
[i
].data
);
1532 #ifdef THREADED_DELTA_SEARCH
1535 * The main thread waits on the condition that (at least) one of the workers
1536 * has stopped working (which is indicated in the .working member of
1537 * struct thread_params).
1538 * When a work thread has completed its work, it sets .working to 0 and
1539 * signals the main thread and waits on the condition that .data_ready
1543 struct thread_params
{
1545 struct object_entry
**list
;
1552 pthread_mutex_t mutex
;
1553 pthread_cond_t cond
;
1554 unsigned *processed
;
1557 static pthread_cond_t progress_cond
= PTHREAD_COND_INITIALIZER
;
1559 static void *threaded_find_deltas(void *arg
)
1561 struct thread_params
*me
= arg
;
1563 while (me
->remaining
) {
1564 find_deltas(me
->list
, &me
->remaining
,
1565 me
->window
, me
->depth
, me
->processed
);
1569 pthread_cond_signal(&progress_cond
);
1573 * We must not set ->data_ready before we wait on the
1574 * condition because the main thread may have set it to 1
1575 * before we get here. In order to be sure that new
1576 * work is available if we see 1 in ->data_ready, it
1577 * was initialized to 0 before this thread was spawned
1578 * and we reset it to 0 right away.
1580 pthread_mutex_lock(&me
->mutex
);
1581 while (!me
->data_ready
)
1582 pthread_cond_wait(&me
->cond
, &me
->mutex
);
1584 pthread_mutex_unlock(&me
->mutex
);
1586 /* leave ->working 1 so that this doesn't get more work assigned */
1590 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
1591 int window
, int depth
, unsigned *processed
)
1593 struct thread_params p
[delta_search_threads
];
1594 int i
, ret
, active_threads
= 0;
1596 if (delta_search_threads
<= 1) {
1597 find_deltas(list
, &list_size
, window
, depth
, processed
);
1601 /* Partition the work amongst work threads. */
1602 for (i
= 0; i
< delta_search_threads
; i
++) {
1603 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
1605 p
[i
].window
= window
;
1607 p
[i
].processed
= processed
;
1609 p
[i
].data_ready
= 0;
1611 /* try to split chunks on "path" boundaries */
1612 while (sub_size
&& sub_size
< list_size
&&
1613 list
[sub_size
]->hash
&&
1614 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
1618 p
[i
].list_size
= sub_size
;
1619 p
[i
].remaining
= sub_size
;
1622 list_size
-= sub_size
;
1625 /* Start work threads. */
1626 for (i
= 0; i
< delta_search_threads
; i
++) {
1627 if (!p
[i
].list_size
)
1629 pthread_mutex_init(&p
[i
].mutex
, NULL
);
1630 pthread_cond_init(&p
[i
].cond
, NULL
);
1631 ret
= pthread_create(&p
[i
].thread
, NULL
,
1632 threaded_find_deltas
, &p
[i
]);
1634 die("unable to create thread: %s", strerror(ret
));
1639 * Now let's wait for work completion. Each time a thread is done
1640 * with its work, we steal half of the remaining work from the
1641 * thread with the largest number of unprocessed objects and give
1642 * it to that newly idle thread. This ensure good load balancing
1643 * until the remaining object list segments are simply too short
1644 * to be worth splitting anymore.
1646 while (active_threads
) {
1647 struct thread_params
*target
= NULL
;
1648 struct thread_params
*victim
= NULL
;
1649 unsigned sub_size
= 0;
1653 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
1658 pthread_cond_wait(&progress_cond
, &progress_mutex
);
1661 for (i
= 0; i
< delta_search_threads
; i
++)
1662 if (p
[i
].remaining
> 2*window
&&
1663 (!victim
|| victim
->remaining
< p
[i
].remaining
))
1666 sub_size
= victim
->remaining
/ 2;
1667 list
= victim
->list
+ victim
->list_size
- sub_size
;
1668 while (sub_size
&& list
[0]->hash
&&
1669 list
[0]->hash
== list
[-1]->hash
) {
1675 * It is possible for some "paths" to have
1676 * so many objects that no hash boundary
1677 * might be found. Let's just steal the
1678 * exact half in that case.
1680 sub_size
= victim
->remaining
/ 2;
1683 target
->list
= list
;
1684 victim
->list_size
-= sub_size
;
1685 victim
->remaining
-= sub_size
;
1687 target
->list_size
= sub_size
;
1688 target
->remaining
= sub_size
;
1689 target
->working
= 1;
1692 pthread_mutex_lock(&target
->mutex
);
1693 target
->data_ready
= 1;
1694 pthread_cond_signal(&target
->cond
);
1695 pthread_mutex_unlock(&target
->mutex
);
1698 pthread_join(target
->thread
, NULL
);
1699 pthread_cond_destroy(&target
->cond
);
1700 pthread_mutex_destroy(&target
->mutex
);
1707 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1710 static int add_ref_tag(const char *path
, const unsigned char *sha1
, int flag
, void *cb_data
)
1712 unsigned char peeled
[20];
1714 if (!prefixcmp(path
, "refs/tags/") && /* is a tag? */
1715 !peel_ref(path
, peeled
) && /* peelable? */
1716 !is_null_sha1(peeled
) && /* annotated tag? */
1717 locate_object_entry(peeled
)) /* object packed? */
1718 add_object_entry(sha1
, OBJ_TAG
, NULL
, 0);
1722 static void prepare_pack(int window
, int depth
)
1724 struct object_entry
**delta_list
;
1725 uint32_t i
, nr_deltas
;
1728 get_object_details();
1731 * If we're locally repacking then we need to be doubly careful
1732 * from now on in order to make sure no stealth corruption gets
1733 * propagated to the new pack. Clients receiving streamed packs
1734 * should validate everything they get anyway so no need to incur
1735 * the additional cost here in that case.
1737 if (!pack_to_stdout
)
1738 do_check_packed_object_crc
= 1;
1740 if (!nr_objects
|| !window
|| !depth
)
1743 delta_list
= xmalloc(nr_objects
* sizeof(*delta_list
));
1746 for (i
= 0; i
< nr_objects
; i
++) {
1747 struct object_entry
*entry
= objects
+ i
;
1750 /* This happens if we decided to reuse existing
1751 * delta from a pack. "reuse_delta &&" is implied.
1755 if (entry
->size
< 50)
1758 if (entry
->no_try_delta
)
1761 if (!entry
->preferred_base
) {
1763 if (entry
->type
< 0)
1764 die("unable to get type of object %s",
1765 sha1_to_hex(entry
->idx
.sha1
));
1767 if (entry
->type
< 0) {
1769 * This object is not found, but we
1770 * don't have to include it anyway.
1776 delta_list
[n
++] = entry
;
1779 if (nr_deltas
&& n
> 1) {
1780 unsigned nr_done
= 0;
1782 progress_state
= start_progress("Compressing objects",
1784 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
1785 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
1786 stop_progress(&progress_state
);
1787 if (nr_done
!= nr_deltas
)
1788 die("inconsistency with delta count");
1793 static int git_pack_config(const char *k
, const char *v
, void *cb
)
1795 if(!strcmp(k
, "pack.window")) {
1796 window
= git_config_int(k
, v
);
1799 if (!strcmp(k
, "pack.windowmemory")) {
1800 window_memory_limit
= git_config_ulong(k
, v
);
1803 if (!strcmp(k
, "pack.depth")) {
1804 depth
= git_config_int(k
, v
);
1807 if (!strcmp(k
, "pack.compression")) {
1808 int level
= git_config_int(k
, v
);
1810 level
= Z_DEFAULT_COMPRESSION
;
1811 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
1812 die("bad pack compression level %d", level
);
1813 pack_compression_level
= level
;
1814 pack_compression_seen
= 1;
1817 if (!strcmp(k
, "pack.deltacachesize")) {
1818 max_delta_cache_size
= git_config_int(k
, v
);
1821 if (!strcmp(k
, "pack.deltacachelimit")) {
1822 cache_max_small_delta_size
= git_config_int(k
, v
);
1825 if (!strcmp(k
, "pack.threads")) {
1826 delta_search_threads
= git_config_int(k
, v
);
1827 if (delta_search_threads
< 0)
1828 die("invalid number of threads specified (%d)",
1829 delta_search_threads
);
1830 #ifndef THREADED_DELTA_SEARCH
1831 if (delta_search_threads
!= 1)
1832 warning("no threads support, ignoring %s", k
);
1836 if (!strcmp(k
, "pack.indexversion")) {
1837 pack_idx_default_version
= git_config_int(k
, v
);
1838 if (pack_idx_default_version
> 2)
1839 die("bad pack.indexversion=%"PRIu32
,
1840 pack_idx_default_version
);
1843 if (!strcmp(k
, "pack.packsizelimit")) {
1844 pack_size_limit_cfg
= git_config_ulong(k
, v
);
1847 return git_default_config(k
, v
, cb
);
1850 static void read_object_list_from_stdin(void)
1852 char line
[40 + 1 + PATH_MAX
+ 2];
1853 unsigned char sha1
[20];
1856 if (!fgets(line
, sizeof(line
), stdin
)) {
1860 die("fgets returned NULL, not EOF, not error!");
1862 die("fgets: %s", strerror(errno
));
1866 if (line
[0] == '-') {
1867 if (get_sha1_hex(line
+1, sha1
))
1868 die("expected edge sha1, got garbage:\n %s",
1870 add_preferred_base(sha1
);
1873 if (get_sha1_hex(line
, sha1
))
1874 die("expected sha1, got garbage:\n %s", line
);
1876 add_preferred_base_object(line
+41);
1877 add_object_entry(sha1
, 0, line
+41, 0);
1881 #define OBJECT_ADDED (1u<<20)
1883 static void show_commit(struct commit
*commit
)
1885 add_object_entry(commit
->object
.sha1
, OBJ_COMMIT
, NULL
, 0);
1886 commit
->object
.flags
|= OBJECT_ADDED
;
1889 static void show_object(struct object_array_entry
*p
)
1891 add_preferred_base_object(p
->name
);
1892 add_object_entry(p
->item
->sha1
, p
->item
->type
, p
->name
, 0);
1893 p
->item
->flags
|= OBJECT_ADDED
;
1896 static void show_edge(struct commit
*commit
)
1898 add_preferred_base(commit
->object
.sha1
);
1901 struct in_pack_object
{
1903 struct object
*object
;
1909 struct in_pack_object
*array
;
1912 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
1914 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->sha1
, p
);
1915 in_pack
->array
[in_pack
->nr
].object
= object
;
1920 * Compare the objects in the offset order, in order to emulate the
1921 * "git rev-list --objects" output that produced the pack originally.
1923 static int ofscmp(const void *a_
, const void *b_
)
1925 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
1926 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
1928 if (a
->offset
< b
->offset
)
1930 else if (a
->offset
> b
->offset
)
1933 return hashcmp(a
->object
->sha1
, b
->object
->sha1
);
1936 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
1938 struct packed_git
*p
;
1939 struct in_pack in_pack
;
1942 memset(&in_pack
, 0, sizeof(in_pack
));
1944 for (p
= packed_git
; p
; p
= p
->next
) {
1945 const unsigned char *sha1
;
1948 for (i
= 0; i
< revs
->num_ignore_packed
; i
++) {
1949 if (matches_pack_name(p
, revs
->ignore_packed
[i
]))
1952 if (revs
->num_ignore_packed
<= i
)
1954 if (open_pack_index(p
))
1955 die("cannot open pack index");
1957 ALLOC_GROW(in_pack
.array
,
1958 in_pack
.nr
+ p
->num_objects
,
1961 for (i
= 0; i
< p
->num_objects
; i
++) {
1962 sha1
= nth_packed_object_sha1(p
, i
);
1963 o
= lookup_unknown_object(sha1
);
1964 if (!(o
->flags
& OBJECT_ADDED
))
1965 mark_in_pack_object(o
, p
, &in_pack
);
1966 o
->flags
|= OBJECT_ADDED
;
1971 qsort(in_pack
.array
, in_pack
.nr
, sizeof(in_pack
.array
[0]),
1973 for (i
= 0; i
< in_pack
.nr
; i
++) {
1974 struct object
*o
= in_pack
.array
[i
].object
;
1975 add_object_entry(o
->sha1
, o
->type
, "", 0);
1978 free(in_pack
.array
);
1981 static void loosen_unused_packed_objects(struct rev_info
*revs
)
1983 struct packed_git
*p
;
1985 const unsigned char *sha1
;
1987 for (p
= packed_git
; p
; p
= p
->next
) {
1988 for (i
= 0; i
< revs
->num_ignore_packed
; i
++) {
1989 if (matches_pack_name(p
, revs
->ignore_packed
[i
]))
1992 if (revs
->num_ignore_packed
<= i
)
1995 if (open_pack_index(p
))
1996 die("cannot open pack index");
1998 for (i
= 0; i
< p
->num_objects
; i
++) {
1999 sha1
= nth_packed_object_sha1(p
, i
);
2000 if (!locate_object_entry(sha1
))
2001 if (force_object_loose(sha1
, p
->mtime
))
2002 die("unable to force loose object");
2007 static void get_object_list(int ac
, const char **av
)
2009 struct rev_info revs
;
2013 init_revisions(&revs
, NULL
);
2014 save_commit_buffer
= 0;
2015 setup_revisions(ac
, av
, &revs
, NULL
);
2017 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2018 int len
= strlen(line
);
2019 if (len
&& line
[len
- 1] == '\n')
2024 if (!strcmp(line
, "--not")) {
2025 flags
^= UNINTERESTING
;
2028 die("not a rev '%s'", line
);
2030 if (handle_revision_arg(line
, &revs
, flags
, 1))
2031 die("bad revision '%s'", line
);
2034 if (prepare_revision_walk(&revs
))
2035 die("revision walk setup failed");
2036 mark_edges_uninteresting(revs
.commits
, &revs
, show_edge
);
2037 traverse_commit_list(&revs
, show_commit
, show_object
);
2039 if (keep_unreachable
)
2040 add_objects_in_unpacked_packs(&revs
);
2041 if (unpack_unreachable
)
2042 loosen_unused_packed_objects(&revs
);
2045 static int adjust_perm(const char *path
, mode_t mode
)
2047 if (chmod(path
, mode
))
2049 return adjust_shared_perm(path
);
2052 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
2054 int use_internal_rev_list
= 0;
2058 int rp_ac_alloc
= 64;
2061 rp_av
= xcalloc(rp_ac_alloc
, sizeof(*rp_av
));
2063 rp_av
[0] = "pack-objects";
2064 rp_av
[1] = "--objects"; /* --thin will make it --objects-edge */
2067 git_config(git_pack_config
, NULL
);
2068 if (!pack_compression_seen
&& core_compression_seen
)
2069 pack_compression_level
= core_compression_level
;
2071 progress
= isatty(2);
2072 for (i
= 1; i
< argc
; i
++) {
2073 const char *arg
= argv
[i
];
2078 if (!strcmp("--non-empty", arg
)) {
2082 if (!strcmp("--local", arg
)) {
2086 if (!strcmp("--incremental", arg
)) {
2090 if (!prefixcmp(arg
, "--compression=")) {
2092 int level
= strtoul(arg
+14, &end
, 0);
2093 if (!arg
[14] || *end
)
2096 level
= Z_DEFAULT_COMPRESSION
;
2097 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
2098 die("bad pack compression level %d", level
);
2099 pack_compression_level
= level
;
2102 if (!prefixcmp(arg
, "--max-pack-size=")) {
2104 pack_size_limit_cfg
= 0;
2105 pack_size_limit
= strtoul(arg
+16, &end
, 0) * 1024 * 1024;
2106 if (!arg
[16] || *end
)
2110 if (!prefixcmp(arg
, "--window=")) {
2112 window
= strtoul(arg
+9, &end
, 0);
2113 if (!arg
[9] || *end
)
2117 if (!prefixcmp(arg
, "--window-memory=")) {
2118 if (!git_parse_ulong(arg
+16, &window_memory_limit
))
2122 if (!prefixcmp(arg
, "--threads=")) {
2124 delta_search_threads
= strtoul(arg
+10, &end
, 0);
2125 if (!arg
[10] || *end
|| delta_search_threads
< 0)
2127 #ifndef THREADED_DELTA_SEARCH
2128 if (delta_search_threads
!= 1)
2129 warning("no threads support, "
2130 "ignoring %s", arg
);
2134 if (!prefixcmp(arg
, "--depth=")) {
2136 depth
= strtoul(arg
+8, &end
, 0);
2137 if (!arg
[8] || *end
)
2141 if (!strcmp("--progress", arg
)) {
2145 if (!strcmp("--all-progress", arg
)) {
2149 if (!strcmp("-q", arg
)) {
2153 if (!strcmp("--no-reuse-delta", arg
)) {
2157 if (!strcmp("--no-reuse-object", arg
)) {
2158 reuse_object
= reuse_delta
= 0;
2161 if (!strcmp("--delta-base-offset", arg
)) {
2162 allow_ofs_delta
= 1;
2165 if (!strcmp("--stdout", arg
)) {
2169 if (!strcmp("--revs", arg
)) {
2170 use_internal_rev_list
= 1;
2173 if (!strcmp("--keep-unreachable", arg
)) {
2174 keep_unreachable
= 1;
2177 if (!strcmp("--unpack-unreachable", arg
)) {
2178 unpack_unreachable
= 1;
2181 if (!strcmp("--include-tag", arg
)) {
2185 if (!strcmp("--unpacked", arg
) ||
2186 !prefixcmp(arg
, "--unpacked=") ||
2187 !strcmp("--reflog", arg
) ||
2188 !strcmp("--all", arg
)) {
2189 use_internal_rev_list
= 1;
2190 if (rp_ac
>= rp_ac_alloc
- 1) {
2191 rp_ac_alloc
= alloc_nr(rp_ac_alloc
);
2192 rp_av
= xrealloc(rp_av
,
2193 rp_ac_alloc
* sizeof(*rp_av
));
2195 rp_av
[rp_ac
++] = arg
;
2198 if (!strcmp("--thin", arg
)) {
2199 use_internal_rev_list
= 1;
2201 rp_av
[1] = "--objects-edge";
2204 if (!prefixcmp(arg
, "--index-version=")) {
2206 pack_idx_default_version
= strtoul(arg
+ 16, &c
, 10);
2207 if (pack_idx_default_version
> 2)
2210 pack_idx_off32_limit
= strtoul(c
+1, &c
, 0);
2211 if (*c
|| pack_idx_off32_limit
& 0x80000000)
2218 /* Traditionally "pack-objects [options] base extra" failed;
2219 * we would however want to take refs parameter that would
2220 * have been given to upstream rev-list ourselves, which means
2221 * we somehow want to say what the base name is. So the
2224 * pack-objects [options] base <refs...>
2226 * in other words, we would treat the first non-option as the
2227 * base_name and send everything else to the internal revision
2231 if (!pack_to_stdout
)
2232 base_name
= argv
[i
++];
2234 if (pack_to_stdout
!= !base_name
)
2237 if (!pack_to_stdout
&& !pack_size_limit
)
2238 pack_size_limit
= pack_size_limit_cfg
;
2240 if (pack_to_stdout
&& pack_size_limit
)
2241 die("--max-pack-size cannot be used to build a pack for transfer.");
2243 if (!pack_to_stdout
&& thin
)
2244 die("--thin cannot be used to build an indexable pack.");
2246 if (keep_unreachable
&& unpack_unreachable
)
2247 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2249 #ifdef THREADED_DELTA_SEARCH
2250 if (!delta_search_threads
) /* --threads=0 means autodetect */
2251 delta_search_threads
= online_cpus();
2254 prepare_packed_git();
2257 progress_state
= start_progress("Counting objects", 0);
2258 if (!use_internal_rev_list
)
2259 read_object_list_from_stdin();
2261 rp_av
[rp_ac
] = NULL
;
2262 get_object_list(rp_ac
, rp_av
);
2264 if (include_tag
&& nr_result
)
2265 for_each_ref(add_ref_tag
, NULL
);
2266 stop_progress(&progress_state
);
2268 if (non_empty
&& !nr_result
)
2271 prepare_pack(window
, depth
);
2274 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
2275 " reused %"PRIu32
" (delta %"PRIu32
")\n",
2276 written
, written_delta
, reused
, reused_delta
);