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 int hash
; /* name hint hash */
48 enum object_type type
;
49 enum object_type in_pack_type
; /* could be delta */
50 unsigned char in_pack_header_size
;
51 unsigned char preferred_base
; /* we do not pack this, but is available
52 * to be used as the base object to delta
55 unsigned char no_try_delta
;
59 * Objects we are going to pack are collected in objects array (dynamically
60 * expanded). nr_objects & nr_alloc controls this array. They are stored
61 * in the order we see -- typically rev-list --objects order that gives us
62 * nice "minimum seek" order.
64 static struct object_entry
*objects
;
65 static struct pack_idx_entry
**written_list
;
66 static uint32_t nr_objects
, nr_alloc
, nr_result
, nr_written
;
69 static int no_reuse_delta
, no_reuse_object
, keep_unreachable
, unpack_unreachable
, include_tag
;
71 static int incremental
;
72 static int allow_ofs_delta
;
73 static const char *base_name
;
74 static int progress
= 1;
75 static int window
= 10;
76 static uint32_t pack_size_limit
, pack_size_limit_cfg
;
77 static int depth
= 50;
78 static int delta_search_threads
= 1;
79 static int pack_to_stdout
;
80 static int num_preferred_base
;
81 static struct progress
*progress_state
;
82 static int pack_compression_level
= Z_DEFAULT_COMPRESSION
;
83 static int pack_compression_seen
;
85 static unsigned long delta_cache_size
= 0;
86 static unsigned long max_delta_cache_size
= 0;
87 static unsigned long cache_max_small_delta_size
= 1000;
89 static unsigned long window_memory_limit
= 0;
92 * The object names in objects array are hashed with this hashtable,
93 * to help looking up the entry by object name.
94 * This hashtable is built after all the objects are seen.
96 static int *object_ix
;
97 static int object_ix_hashsz
;
102 static uint32_t written
, written_delta
;
103 static uint32_t reused
, reused_delta
;
106 static void *delta_against(void *buf
, unsigned long size
, struct object_entry
*entry
)
108 unsigned long othersize
, delta_size
;
109 enum object_type type
;
110 void *otherbuf
= read_sha1_file(entry
->delta
->idx
.sha1
, &type
, &othersize
);
114 die("unable to read %s", sha1_to_hex(entry
->delta
->idx
.sha1
));
115 delta_buf
= diff_delta(otherbuf
, othersize
,
116 buf
, size
, &delta_size
, 0);
117 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
118 die("delta size changed");
125 * The per-object header is a pretty dense thing, which is
126 * - first byte: low four bits are "size", then three bits of "type",
127 * and the high bit is "size continues".
128 * - each byte afterwards: low seven bits are size continuation,
129 * with the high bit being "size continues"
131 static int encode_header(enum object_type type
, unsigned long size
, unsigned char *hdr
)
136 if (type
< OBJ_COMMIT
|| type
> OBJ_REF_DELTA
)
137 die("bad type %d", type
);
139 c
= (type
<< 4) | (size
& 15);
152 * we are going to reuse the existing object data as is. make
153 * sure it is not corrupt.
155 static int check_pack_inflate(struct packed_git
*p
,
156 struct pack_window
**w_curs
,
159 unsigned long expect
)
162 unsigned char fakebuf
[4096], *in
;
165 memset(&stream
, 0, sizeof(stream
));
166 inflateInit(&stream
);
168 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
170 stream
.next_out
= fakebuf
;
171 stream
.avail_out
= sizeof(fakebuf
);
172 st
= inflate(&stream
, Z_FINISH
);
173 offset
+= stream
.next_in
- in
;
174 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
176 return (st
== Z_STREAM_END
&&
177 stream
.total_out
== expect
&&
178 stream
.total_in
== len
) ? 0 : -1;
181 static int check_pack_crc(struct packed_git
*p
, struct pack_window
**w_curs
,
182 off_t offset
, off_t len
, unsigned int nr
)
184 const uint32_t *index_crc
;
185 uint32_t data_crc
= crc32(0, Z_NULL
, 0);
189 void *data
= use_pack(p
, w_curs
, offset
, &avail
);
192 data_crc
= crc32(data_crc
, data
, avail
);
197 index_crc
= p
->index_data
;
198 index_crc
+= 2 + 256 + p
->num_objects
* (20/4) + nr
;
200 return data_crc
!= ntohl(*index_crc
);
203 static void copy_pack_data(struct sha1file
*f
,
204 struct packed_git
*p
,
205 struct pack_window
**w_curs
,
213 in
= use_pack(p
, w_curs
, offset
, &avail
);
215 avail
= (unsigned int)len
;
216 sha1write(f
, in
, avail
);
222 static unsigned long write_object(struct sha1file
*f
,
223 struct object_entry
*entry
,
227 enum object_type type
;
229 unsigned char header
[10];
230 unsigned char dheader
[10];
233 enum object_type obj_type
;
235 /* write limit if limited packsize and not first object */
236 unsigned long limit
= pack_size_limit
&& nr_written
?
237 pack_size_limit
- write_offset
: 0;
239 int usable_delta
= !entry
->delta
? 0 :
240 /* yes if unlimited packfile */
241 !pack_size_limit
? 1 :
242 /* no if base written to previous pack */
243 entry
->delta
->idx
.offset
== (off_t
)-1 ? 0 :
244 /* otherwise double-check written to this
245 * pack, like we do below
247 entry
->delta
->idx
.offset
? 1 : 0;
252 obj_type
= entry
->type
;
254 to_reuse
= 0; /* explicit */
255 else if (!entry
->in_pack
)
256 to_reuse
= 0; /* can't reuse what we don't have */
257 else if (obj_type
== OBJ_REF_DELTA
|| obj_type
== OBJ_OFS_DELTA
)
258 /* check_object() decided it for us ... */
259 to_reuse
= usable_delta
;
260 /* ... but pack split may override that */
261 else if (obj_type
!= entry
->in_pack_type
)
262 to_reuse
= 0; /* pack has delta which is unusable */
263 else if (entry
->delta
)
264 to_reuse
= 0; /* we want to pack afresh */
266 to_reuse
= 1; /* we have it in-pack undeltified,
267 * and we do not need to deltify it.
272 unsigned long maxsize
;
275 buf
= read_sha1_file(entry
->idx
.sha1
, &obj_type
, &size
);
277 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
278 } else if (entry
->delta_data
) {
279 size
= entry
->delta_size
;
280 buf
= entry
->delta_data
;
281 entry
->delta_data
= NULL
;
282 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
283 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
285 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
287 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
288 buf
= delta_against(buf
, size
, entry
);
289 size
= entry
->delta_size
;
290 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
291 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
293 /* compress the data to store and put compressed length in datalen */
294 memset(&stream
, 0, sizeof(stream
));
295 deflateInit(&stream
, pack_compression_level
);
296 maxsize
= deflateBound(&stream
, size
);
297 out
= xmalloc(maxsize
);
299 stream
.next_in
= buf
;
300 stream
.avail_in
= size
;
301 stream
.next_out
= out
;
302 stream
.avail_out
= maxsize
;
303 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
306 datalen
= stream
.total_out
;
309 * The object header is a byte of 'type' followed by zero or
310 * more bytes of length.
312 hdrlen
= encode_header(obj_type
, size
, header
);
314 if (obj_type
== OBJ_OFS_DELTA
) {
316 * Deltas with relative base contain an additional
317 * encoding of the relative offset for the delta
318 * base from this object's position in the pack.
320 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
321 unsigned pos
= sizeof(dheader
) - 1;
322 dheader
[pos
] = ofs
& 127;
324 dheader
[--pos
] = 128 | (--ofs
& 127);
325 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
330 sha1write(f
, header
, hdrlen
);
331 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
332 hdrlen
+= sizeof(dheader
) - pos
;
333 } else if (obj_type
== OBJ_REF_DELTA
) {
335 * Deltas with a base reference contain
336 * an additional 20 bytes for the base sha1.
338 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
) {
352 sha1write(f
, header
, hdrlen
);
354 sha1write(f
, out
, datalen
);
359 struct packed_git
*p
= entry
->in_pack
;
360 struct pack_window
*w_curs
= NULL
;
361 struct revindex_entry
*revidx
;
365 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
366 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
369 hdrlen
= encode_header(obj_type
, entry
->size
, header
);
370 offset
= entry
->in_pack_offset
;
371 revidx
= find_pack_revindex(p
, offset
);
372 datalen
= revidx
[1].offset
- offset
;
373 if (!pack_to_stdout
&& p
->index_version
> 1 &&
374 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
))
375 die("bad packed object CRC for %s", sha1_to_hex(entry
->idx
.sha1
));
376 offset
+= entry
->in_pack_header_size
;
377 datalen
-= entry
->in_pack_header_size
;
378 if (obj_type
== OBJ_OFS_DELTA
) {
379 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
380 unsigned pos
= sizeof(dheader
) - 1;
381 dheader
[pos
] = ofs
& 127;
383 dheader
[--pos
] = 128 | (--ofs
& 127);
384 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
)
386 sha1write(f
, header
, hdrlen
);
387 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
388 hdrlen
+= sizeof(dheader
) - pos
;
389 } else if (obj_type
== OBJ_REF_DELTA
) {
390 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
)
392 sha1write(f
, header
, hdrlen
);
393 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
396 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
)
398 sha1write(f
, header
, hdrlen
);
401 if (!pack_to_stdout
&& p
->index_version
== 1 &&
402 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry
->size
))
403 die("corrupt packed object for %s", sha1_to_hex(entry
->idx
.sha1
));
404 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
412 entry
->idx
.crc32
= crc32_end(f
);
413 return hdrlen
+ datalen
;
416 static off_t
write_one(struct sha1file
*f
,
417 struct object_entry
*e
,
422 /* offset is non zero if object is written already. */
423 if (e
->idx
.offset
|| e
->preferred_base
)
426 /* if we are deltified, write out base object first. */
428 offset
= write_one(f
, e
->delta
, offset
);
433 e
->idx
.offset
= offset
;
434 size
= write_object(f
, e
, offset
);
439 written_list
[nr_written
++] = &e
->idx
;
441 /* make sure off_t is sufficiently large not to wrap */
442 if (offset
> offset
+ size
)
443 die("pack too large for current definition of off_t");
444 return offset
+ size
;
447 /* forward declaration for write_pack_file */
448 static int adjust_perm(const char *path
, mode_t mode
);
450 static void write_pack_file(void)
454 off_t offset
, offset_one
, last_obj_offset
= 0;
455 struct pack_header hdr
;
456 int do_progress
= progress
>> pack_to_stdout
;
457 uint32_t nr_remaining
= nr_result
;
458 time_t last_mtime
= 0;
461 progress_state
= start_progress("Writing objects", nr_result
);
462 written_list
= xmalloc(nr_objects
* sizeof(*written_list
));
465 unsigned char sha1
[20];
466 char *pack_tmp_name
= NULL
;
468 if (pack_to_stdout
) {
469 f
= sha1fd_throughput(1, "<stdout>", progress_state
);
471 char tmpname
[PATH_MAX
];
473 snprintf(tmpname
, sizeof(tmpname
),
474 "%s/tmp_pack_XXXXXX", get_object_directory());
475 fd
= xmkstemp(tmpname
);
476 pack_tmp_name
= xstrdup(tmpname
);
477 f
= sha1fd(fd
, pack_tmp_name
);
480 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
481 hdr
.hdr_version
= htonl(PACK_VERSION
);
482 hdr
.hdr_entries
= htonl(nr_remaining
);
483 sha1write(f
, &hdr
, sizeof(hdr
));
484 offset
= sizeof(hdr
);
486 for (; i
< nr_objects
; i
++) {
487 last_obj_offset
= offset
;
488 offset_one
= write_one(f
, objects
+ i
, offset
);
492 display_progress(progress_state
, written
);
496 * Did we write the wrong # entries in the header?
497 * If so, rewrite it like in fast-import
499 if (pack_to_stdout
|| nr_written
== nr_remaining
) {
500 sha1close(f
, sha1
, 1);
502 int fd
= sha1close(f
, NULL
, 0);
503 fixup_pack_header_footer(fd
, sha1
, pack_tmp_name
, nr_written
);
507 if (!pack_to_stdout
) {
508 mode_t mode
= umask(0);
510 char *idx_tmp_name
, tmpname
[PATH_MAX
];
515 idx_tmp_name
= write_idx_file(NULL
, written_list
,
518 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.pack",
519 base_name
, sha1_to_hex(sha1
));
520 if (adjust_perm(pack_tmp_name
, mode
))
521 die("unable to make temporary pack file readable: %s",
523 if (rename(pack_tmp_name
, tmpname
))
524 die("unable to rename temporary pack file: %s",
528 * Packs are runtime accessed in their mtime
529 * order since newer packs are more likely to contain
530 * younger objects. So if we are creating multiple
531 * packs then we should modify the mtime of later ones
532 * to preserve this property.
534 if (stat(tmpname
, &st
) < 0) {
535 warning("failed to stat %s: %s",
536 tmpname
, strerror(errno
));
537 } else if (!last_mtime
) {
538 last_mtime
= st
.st_mtime
;
541 utb
.actime
= st
.st_atime
;
542 utb
.modtime
= --last_mtime
;
543 if (utime(tmpname
, &utb
) < 0)
544 warning("failed utime() on %s: %s",
545 tmpname
, strerror(errno
));
548 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.idx",
549 base_name
, sha1_to_hex(sha1
));
550 if (adjust_perm(idx_tmp_name
, mode
))
551 die("unable to make temporary index file readable: %s",
553 if (rename(idx_tmp_name
, tmpname
))
554 die("unable to rename temporary index file: %s",
559 puts(sha1_to_hex(sha1
));
562 /* mark written objects as written to previous pack */
563 for (j
= 0; j
< nr_written
; j
++) {
564 written_list
[j
]->offset
= (off_t
)-1;
566 nr_remaining
-= nr_written
;
567 } while (nr_remaining
&& i
< nr_objects
);
570 stop_progress(&progress_state
);
571 if (written
!= nr_result
)
572 die("wrote %u objects while expecting %u", written
, nr_result
);
574 * We have scanned through [0 ... i). Since we have written
575 * the correct number of objects, the remaining [i ... nr_objects)
576 * items must be either already written (due to out-of-order delta base)
577 * or a preferred base. Count those which are neither and complain if any.
579 for (j
= 0; i
< nr_objects
; i
++) {
580 struct object_entry
*e
= objects
+ i
;
581 j
+= !e
->idx
.offset
&& !e
->preferred_base
;
584 die("wrote %u objects as expected but %u unwritten", written
, j
);
587 static int locate_object_entry_hash(const unsigned char *sha1
)
591 memcpy(&ui
, sha1
, sizeof(unsigned int));
592 i
= ui
% object_ix_hashsz
;
593 while (0 < object_ix
[i
]) {
594 if (!hashcmp(sha1
, objects
[object_ix
[i
] - 1].idx
.sha1
))
596 if (++i
== object_ix_hashsz
)
602 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
606 if (!object_ix_hashsz
)
609 i
= locate_object_entry_hash(sha1
);
611 return &objects
[object_ix
[i
]-1];
615 static void rehash_objects(void)
618 struct object_entry
*oe
;
620 object_ix_hashsz
= nr_objects
* 3;
621 if (object_ix_hashsz
< 1024)
622 object_ix_hashsz
= 1024;
623 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
624 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
625 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
626 int ix
= locate_object_entry_hash(oe
->idx
.sha1
);
630 object_ix
[ix
] = i
+ 1;
634 static unsigned name_hash(const char *name
)
643 * This effectively just creates a sortable number from the
644 * last sixteen non-whitespace characters. Last characters
645 * count "most", so things that end in ".c" sort together.
647 while ((c
= *name
++) != 0) {
650 hash
= (hash
>> 2) + (c
<< 24);
655 static void setup_delta_attr_check(struct git_attr_check
*check
)
657 static struct git_attr
*attr_delta
;
660 attr_delta
= git_attr("delta", 5);
662 check
[0].attr
= attr_delta
;
665 static int no_try_delta(const char *path
)
667 struct git_attr_check check
[1];
669 setup_delta_attr_check(check
);
670 if (git_checkattr(path
, ARRAY_SIZE(check
), check
))
672 if (ATTR_FALSE(check
->value
))
677 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
678 const char *name
, int exclude
)
680 struct object_entry
*entry
;
681 struct packed_git
*p
, *found_pack
= NULL
;
682 off_t found_offset
= 0;
684 unsigned hash
= name_hash(name
);
686 ix
= nr_objects
? locate_object_entry_hash(sha1
) : -1;
689 entry
= objects
+ object_ix
[ix
] - 1;
690 if (!entry
->preferred_base
)
692 entry
->preferred_base
= 1;
697 for (p
= packed_git
; p
; p
= p
->next
) {
698 off_t offset
= find_pack_entry_one(sha1
, p
);
701 found_offset
= offset
;
708 if (local
&& !p
->pack_local
)
713 if (nr_objects
>= nr_alloc
) {
714 nr_alloc
= (nr_alloc
+ 1024) * 3 / 2;
715 objects
= xrealloc(objects
, nr_alloc
* sizeof(*entry
));
718 entry
= objects
+ nr_objects
++;
719 memset(entry
, 0, sizeof(*entry
));
720 hashcpy(entry
->idx
.sha1
, sha1
);
725 entry
->preferred_base
= 1;
729 entry
->in_pack
= found_pack
;
730 entry
->in_pack_offset
= found_offset
;
733 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
736 object_ix
[-1 - ix
] = nr_objects
;
738 display_progress(progress_state
, nr_objects
);
740 if (name
&& no_try_delta(name
))
741 entry
->no_try_delta
= 1;
746 struct pbase_tree_cache
{
747 unsigned char sha1
[20];
751 unsigned long tree_size
;
754 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
755 static int pbase_tree_cache_ix(const unsigned char *sha1
)
757 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
759 static int pbase_tree_cache_ix_incr(int ix
)
761 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
764 static struct pbase_tree
{
765 struct pbase_tree
*next
;
766 /* This is a phony "cache" entry; we are not
767 * going to evict it nor find it through _get()
768 * mechanism -- this is for the toplevel node that
769 * would almost always change with any commit.
771 struct pbase_tree_cache pcache
;
774 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
776 struct pbase_tree_cache
*ent
, *nent
;
779 enum object_type type
;
781 int my_ix
= pbase_tree_cache_ix(sha1
);
782 int available_ix
= -1;
784 /* pbase-tree-cache acts as a limited hashtable.
785 * your object will be found at your index or within a few
786 * slots after that slot if it is cached.
788 for (neigh
= 0; neigh
< 8; neigh
++) {
789 ent
= pbase_tree_cache
[my_ix
];
790 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
794 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
795 ((0 <= available_ix
) &&
796 (!ent
&& pbase_tree_cache
[available_ix
])))
797 available_ix
= my_ix
;
800 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
803 /* Did not find one. Either we got a bogus request or
804 * we need to read and perhaps cache.
806 data
= read_sha1_file(sha1
, &type
, &size
);
809 if (type
!= OBJ_TREE
) {
814 /* We need to either cache or return a throwaway copy */
816 if (available_ix
< 0)
819 ent
= pbase_tree_cache
[available_ix
];
820 my_ix
= available_ix
;
824 nent
= xmalloc(sizeof(*nent
));
825 nent
->temporary
= (available_ix
< 0);
828 /* evict and reuse */
829 free(ent
->tree_data
);
832 hashcpy(nent
->sha1
, sha1
);
833 nent
->tree_data
= data
;
834 nent
->tree_size
= size
;
836 if (!nent
->temporary
)
837 pbase_tree_cache
[my_ix
] = nent
;
841 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
843 if (!cache
->temporary
) {
847 free(cache
->tree_data
);
851 static int name_cmp_len(const char *name
)
854 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
859 static void add_pbase_object(struct tree_desc
*tree
,
862 const char *fullname
)
864 struct name_entry entry
;
867 while (tree_entry(tree
,&entry
)) {
868 if (S_ISGITLINK(entry
.mode
))
870 cmp
= tree_entry_len(entry
.path
, entry
.sha1
) != cmplen
? 1 :
871 memcmp(name
, entry
.path
, cmplen
);
876 if (name
[cmplen
] != '/') {
877 add_object_entry(entry
.sha1
,
878 object_type(entry
.mode
),
882 if (S_ISDIR(entry
.mode
)) {
883 struct tree_desc sub
;
884 struct pbase_tree_cache
*tree
;
885 const char *down
= name
+cmplen
+1;
886 int downlen
= name_cmp_len(down
);
888 tree
= pbase_tree_get(entry
.sha1
);
891 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
893 add_pbase_object(&sub
, down
, downlen
, fullname
);
894 pbase_tree_put(tree
);
899 static unsigned *done_pbase_paths
;
900 static int done_pbase_paths_num
;
901 static int done_pbase_paths_alloc
;
902 static int done_pbase_path_pos(unsigned hash
)
905 int hi
= done_pbase_paths_num
;
907 int mi
= (hi
+ lo
) / 2;
908 if (done_pbase_paths
[mi
] == hash
)
910 if (done_pbase_paths
[mi
] < hash
)
918 static int check_pbase_path(unsigned hash
)
920 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
924 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
925 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
926 done_pbase_paths
= xrealloc(done_pbase_paths
,
927 done_pbase_paths_alloc
*
930 done_pbase_paths_num
++;
931 if (pos
< done_pbase_paths_num
)
932 memmove(done_pbase_paths
+ pos
+ 1,
933 done_pbase_paths
+ pos
,
934 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
935 done_pbase_paths
[pos
] = hash
;
939 static void add_preferred_base_object(const char *name
)
941 struct pbase_tree
*it
;
943 unsigned hash
= name_hash(name
);
945 if (!num_preferred_base
|| check_pbase_path(hash
))
948 cmplen
= name_cmp_len(name
);
949 for (it
= pbase_tree
; it
; it
= it
->next
) {
951 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
954 struct tree_desc tree
;
955 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
956 add_pbase_object(&tree
, name
, cmplen
, name
);
961 static void add_preferred_base(unsigned char *sha1
)
963 struct pbase_tree
*it
;
966 unsigned char tree_sha1
[20];
968 if (window
<= num_preferred_base
++)
971 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
975 for (it
= pbase_tree
; it
; it
= it
->next
) {
976 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
982 it
= xcalloc(1, sizeof(*it
));
983 it
->next
= pbase_tree
;
986 hashcpy(it
->pcache
.sha1
, tree_sha1
);
987 it
->pcache
.tree_data
= data
;
988 it
->pcache
.tree_size
= size
;
991 static void check_object(struct object_entry
*entry
)
993 if (entry
->in_pack
) {
994 struct packed_git
*p
= entry
->in_pack
;
995 struct pack_window
*w_curs
= NULL
;
996 const unsigned char *base_ref
= NULL
;
997 struct object_entry
*base_entry
;
998 unsigned long used
, used_0
;
1001 unsigned char *buf
, c
;
1003 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1006 * We want in_pack_type even if we do not reuse delta
1007 * since non-delta representations could still be reused.
1009 used
= unpack_object_header_gently(buf
, avail
,
1010 &entry
->in_pack_type
,
1014 * Determine if this is a delta and if so whether we can
1015 * reuse it or not. Otherwise let's find out as cheaply as
1016 * possible what the actual type and size for this object is.
1018 switch (entry
->in_pack_type
) {
1020 /* Not a delta hence we've already got all we need. */
1021 entry
->type
= entry
->in_pack_type
;
1022 entry
->in_pack_header_size
= used
;
1023 unuse_pack(&w_curs
);
1026 if (!no_reuse_delta
&& !entry
->preferred_base
)
1027 base_ref
= use_pack(p
, &w_curs
,
1028 entry
->in_pack_offset
+ used
, NULL
);
1029 entry
->in_pack_header_size
= used
+ 20;
1032 buf
= use_pack(p
, &w_curs
,
1033 entry
->in_pack_offset
+ used
, NULL
);
1039 if (!ofs
|| MSB(ofs
, 7))
1040 die("delta base offset overflow in pack for %s",
1041 sha1_to_hex(entry
->idx
.sha1
));
1043 ofs
= (ofs
<< 7) + (c
& 127);
1045 if (ofs
>= entry
->in_pack_offset
)
1046 die("delta base offset out of bound for %s",
1047 sha1_to_hex(entry
->idx
.sha1
));
1048 ofs
= entry
->in_pack_offset
- ofs
;
1049 if (!no_reuse_delta
&& !entry
->preferred_base
) {
1050 struct revindex_entry
*revidx
;
1051 revidx
= find_pack_revindex(p
, ofs
);
1052 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1054 entry
->in_pack_header_size
= used
+ used_0
;
1058 if (base_ref
&& (base_entry
= locate_object_entry(base_ref
))) {
1060 * If base_ref was set above that means we wish to
1061 * reuse delta data, and we even found that base
1062 * in the list of objects we want to pack. Goodie!
1064 * Depth value does not matter - find_deltas() will
1065 * never consider reused delta as the base object to
1066 * deltify other objects against, in order to avoid
1069 entry
->type
= entry
->in_pack_type
;
1070 entry
->delta
= base_entry
;
1071 entry
->delta_sibling
= base_entry
->delta_child
;
1072 base_entry
->delta_child
= entry
;
1073 unuse_pack(&w_curs
);
1079 * This must be a delta and we already know what the
1080 * final object type is. Let's extract the actual
1081 * object size from the delta header.
1083 entry
->size
= get_size_from_delta(p
, &w_curs
,
1084 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1085 unuse_pack(&w_curs
);
1090 * No choice but to fall back to the recursive delta walk
1091 * with sha1_object_info() to find about the object type
1094 unuse_pack(&w_curs
);
1097 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1098 if (entry
->type
< 0)
1099 die("unable to get type of object %s",
1100 sha1_to_hex(entry
->idx
.sha1
));
1103 static int pack_offset_sort(const void *_a
, const void *_b
)
1105 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1106 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1108 /* avoid filesystem trashing with loose objects */
1109 if (!a
->in_pack
&& !b
->in_pack
)
1110 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1112 if (a
->in_pack
< b
->in_pack
)
1114 if (a
->in_pack
> b
->in_pack
)
1116 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1117 (a
->in_pack_offset
> b
->in_pack_offset
);
1120 static void get_object_details(void)
1123 struct object_entry
**sorted_by_offset
;
1125 sorted_by_offset
= xcalloc(nr_objects
, sizeof(struct object_entry
*));
1126 for (i
= 0; i
< nr_objects
; i
++)
1127 sorted_by_offset
[i
] = objects
+ i
;
1128 qsort(sorted_by_offset
, nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1130 init_pack_revindex();
1132 for (i
= 0; i
< nr_objects
; i
++)
1133 check_object(sorted_by_offset
[i
]);
1135 free(sorted_by_offset
);
1139 * We search for deltas in a list sorted by type, by filename hash, and then
1140 * by size, so that we see progressively smaller and smaller files.
1141 * That's because we prefer deltas to be from the bigger file
1142 * to the smaller -- deletes are potentially cheaper, but perhaps
1143 * more importantly, the bigger file is likely the more recent
1144 * one. The deepest deltas are therefore the oldest objects which are
1145 * less susceptible to be accessed often.
1147 static int type_size_sort(const void *_a
, const void *_b
)
1149 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1150 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1152 if (a
->type
> b
->type
)
1154 if (a
->type
< b
->type
)
1156 if (a
->hash
> b
->hash
)
1158 if (a
->hash
< b
->hash
)
1160 if (a
->preferred_base
> b
->preferred_base
)
1162 if (a
->preferred_base
< b
->preferred_base
)
1164 if (a
->size
> b
->size
)
1166 if (a
->size
< b
->size
)
1168 return a
< b
? -1 : (a
> b
); /* newest first */
1172 struct object_entry
*entry
;
1174 struct delta_index
*index
;
1178 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1179 unsigned long delta_size
)
1181 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1184 if (delta_size
< cache_max_small_delta_size
)
1187 /* cache delta, if objects are large enough compared to delta size */
1188 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1194 #ifdef THREADED_DELTA_SEARCH
1196 static pthread_mutex_t read_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1197 #define read_lock() pthread_mutex_lock(&read_mutex)
1198 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1200 static pthread_mutex_t cache_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1201 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1202 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1204 static pthread_mutex_t progress_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1205 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1206 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1210 #define read_lock() (void)0
1211 #define read_unlock() (void)0
1212 #define cache_lock() (void)0
1213 #define cache_unlock() (void)0
1214 #define progress_lock() (void)0
1215 #define progress_unlock() (void)0
1219 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1220 unsigned max_depth
, unsigned long *mem_usage
)
1222 struct object_entry
*trg_entry
= trg
->entry
;
1223 struct object_entry
*src_entry
= src
->entry
;
1224 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1226 enum object_type type
;
1229 /* Don't bother doing diffs between different types */
1230 if (trg_entry
->type
!= src_entry
->type
)
1234 * We do not bother to try a delta that we discarded
1235 * on an earlier try, but only when reusing delta data.
1237 if (!no_reuse_delta
&& trg_entry
->in_pack
&&
1238 trg_entry
->in_pack
== src_entry
->in_pack
&&
1239 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1240 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1243 /* Let's not bust the allowed depth. */
1244 if (src
->depth
>= max_depth
)
1247 /* Now some size filtering heuristics. */
1248 trg_size
= trg_entry
->size
;
1249 if (!trg_entry
->delta
) {
1250 max_size
= trg_size
/2 - 20;
1253 max_size
= trg_entry
->delta_size
;
1254 ref_depth
= trg
->depth
;
1256 max_size
= max_size
* (max_depth
- src
->depth
) /
1257 (max_depth
- ref_depth
+ 1);
1260 src_size
= src_entry
->size
;
1261 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1262 if (sizediff
>= max_size
)
1264 if (trg_size
< src_size
/ 32)
1267 /* Load data if not already done */
1270 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1273 die("object %s cannot be read",
1274 sha1_to_hex(trg_entry
->idx
.sha1
));
1276 die("object %s inconsistent object length (%lu vs %lu)",
1277 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1282 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1285 die("object %s cannot be read",
1286 sha1_to_hex(src_entry
->idx
.sha1
));
1288 die("object %s inconsistent object length (%lu vs %lu)",
1289 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1293 src
->index
= create_delta_index(src
->data
, src_size
);
1295 static int warned
= 0;
1297 warning("suboptimal pack - out of memory");
1300 *mem_usage
+= sizeof_delta_index(src
->index
);
1303 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1307 if (trg_entry
->delta
) {
1308 /* Prefer only shallower same-sized deltas. */
1309 if (delta_size
== trg_entry
->delta_size
&&
1310 src
->depth
+ 1 >= trg
->depth
) {
1317 * Handle memory allocation outside of the cache
1318 * accounting lock. Compiler will optimize the strangeness
1319 * away when THREADED_DELTA_SEARCH is not defined.
1321 free(trg_entry
->delta_data
);
1323 if (trg_entry
->delta_data
) {
1324 delta_cache_size
-= trg_entry
->delta_size
;
1325 trg_entry
->delta_data
= NULL
;
1327 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1328 delta_cache_size
+= delta_size
;
1330 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1336 trg_entry
->delta
= src_entry
;
1337 trg_entry
->delta_size
= delta_size
;
1338 trg
->depth
= src
->depth
+ 1;
1343 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1345 struct object_entry
*child
= me
->delta_child
;
1348 unsigned int c
= check_delta_limit(child
, n
+ 1);
1351 child
= child
->delta_sibling
;
1356 static unsigned long free_unpacked(struct unpacked
*n
)
1358 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1359 free_delta_index(n
->index
);
1362 freed_mem
+= n
->entry
->size
;
1371 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1372 int window
, int depth
, unsigned *processed
)
1374 uint32_t i
, idx
= 0, count
= 0;
1375 unsigned int array_size
= window
* sizeof(struct unpacked
);
1376 struct unpacked
*array
;
1377 unsigned long mem_usage
= 0;
1379 array
= xmalloc(array_size
);
1380 memset(array
, 0, array_size
);
1383 struct object_entry
*entry
= *list
++;
1384 struct unpacked
*n
= array
+ idx
;
1385 int j
, max_depth
, best_base
= -1;
1393 if (!entry
->preferred_base
) {
1395 display_progress(progress_state
, *processed
);
1399 mem_usage
-= free_unpacked(n
);
1402 while (window_memory_limit
&&
1403 mem_usage
> window_memory_limit
&&
1405 uint32_t tail
= (idx
+ window
- count
) % window
;
1406 mem_usage
-= free_unpacked(array
+ tail
);
1410 /* We do not compute delta to *create* objects we are not
1413 if (entry
->preferred_base
)
1417 * If the current object is at pack edge, take the depth the
1418 * objects that depend on the current object into account
1419 * otherwise they would become too deep.
1422 if (entry
->delta_child
) {
1423 max_depth
-= check_delta_limit(entry
, 0);
1431 uint32_t other_idx
= idx
+ j
;
1433 if (other_idx
>= window
)
1434 other_idx
-= window
;
1435 m
= array
+ other_idx
;
1438 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1442 best_base
= other_idx
;
1445 /* if we made n a delta, and if n is already at max
1446 * depth, leaving it in the window is pointless. we
1447 * should evict it first.
1449 if (entry
->delta
&& depth
<= n
->depth
)
1453 * Move the best delta base up in the window, after the
1454 * currently deltified object, to keep it longer. It will
1455 * be the first base object to be attempted next.
1458 struct unpacked swap
= array
[best_base
];
1459 int dist
= (window
+ idx
- best_base
) % window
;
1460 int dst
= best_base
;
1462 int src
= (dst
+ 1) % window
;
1463 array
[dst
] = array
[src
];
1471 if (count
+ 1 < window
)
1477 for (i
= 0; i
< window
; ++i
) {
1478 free_delta_index(array
[i
].index
);
1479 free(array
[i
].data
);
1484 #ifdef THREADED_DELTA_SEARCH
1487 * The main thread waits on the condition that (at least) one of the workers
1488 * has stopped working (which is indicated in the .working member of
1489 * struct thread_params).
1490 * When a work thread has completed its work, it sets .working to 0 and
1491 * signals the main thread and waits on the condition that .data_ready
1495 struct thread_params
{
1497 struct object_entry
**list
;
1504 pthread_mutex_t mutex
;
1505 pthread_cond_t cond
;
1506 unsigned *processed
;
1509 static pthread_cond_t progress_cond
= PTHREAD_COND_INITIALIZER
;
1511 static void *threaded_find_deltas(void *arg
)
1513 struct thread_params
*me
= arg
;
1515 while (me
->remaining
) {
1516 find_deltas(me
->list
, &me
->remaining
,
1517 me
->window
, me
->depth
, me
->processed
);
1521 pthread_cond_signal(&progress_cond
);
1525 * We must not set ->data_ready before we wait on the
1526 * condition because the main thread may have set it to 1
1527 * before we get here. In order to be sure that new
1528 * work is available if we see 1 in ->data_ready, it
1529 * was initialized to 0 before this thread was spawned
1530 * and we reset it to 0 right away.
1532 pthread_mutex_lock(&me
->mutex
);
1533 while (!me
->data_ready
)
1534 pthread_cond_wait(&me
->cond
, &me
->mutex
);
1536 pthread_mutex_unlock(&me
->mutex
);
1538 /* leave ->working 1 so that this doesn't get more work assigned */
1542 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
1543 int window
, int depth
, unsigned *processed
)
1545 struct thread_params p
[delta_search_threads
];
1546 int i
, ret
, active_threads
= 0;
1548 if (delta_search_threads
<= 1) {
1549 find_deltas(list
, &list_size
, window
, depth
, processed
);
1553 /* Partition the work amongst work threads. */
1554 for (i
= 0; i
< delta_search_threads
; i
++) {
1555 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
1557 p
[i
].window
= window
;
1559 p
[i
].processed
= processed
;
1561 p
[i
].data_ready
= 0;
1563 /* try to split chunks on "path" boundaries */
1564 while (sub_size
&& sub_size
< list_size
&&
1565 list
[sub_size
]->hash
&&
1566 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
1570 p
[i
].list_size
= sub_size
;
1571 p
[i
].remaining
= sub_size
;
1574 list_size
-= sub_size
;
1577 /* Start work threads. */
1578 for (i
= 0; i
< delta_search_threads
; i
++) {
1579 if (!p
[i
].list_size
)
1581 pthread_mutex_init(&p
[i
].mutex
, NULL
);
1582 pthread_cond_init(&p
[i
].cond
, NULL
);
1583 ret
= pthread_create(&p
[i
].thread
, NULL
,
1584 threaded_find_deltas
, &p
[i
]);
1586 die("unable to create thread: %s", strerror(ret
));
1591 * Now let's wait for work completion. Each time a thread is done
1592 * with its work, we steal half of the remaining work from the
1593 * thread with the largest number of unprocessed objects and give
1594 * it to that newly idle thread. This ensure good load balancing
1595 * until the remaining object list segments are simply too short
1596 * to be worth splitting anymore.
1598 while (active_threads
) {
1599 struct thread_params
*target
= NULL
;
1600 struct thread_params
*victim
= NULL
;
1601 unsigned sub_size
= 0;
1605 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
1610 pthread_cond_wait(&progress_cond
, &progress_mutex
);
1613 for (i
= 0; i
< delta_search_threads
; i
++)
1614 if (p
[i
].remaining
> 2*window
&&
1615 (!victim
|| victim
->remaining
< p
[i
].remaining
))
1618 sub_size
= victim
->remaining
/ 2;
1619 list
= victim
->list
+ victim
->list_size
- sub_size
;
1620 while (sub_size
&& list
[0]->hash
&&
1621 list
[0]->hash
== list
[-1]->hash
) {
1627 * It is possible for some "paths" to have
1628 * so many objects that no hash boundary
1629 * might be found. Let's just steal the
1630 * exact half in that case.
1632 sub_size
= victim
->remaining
/ 2;
1635 target
->list
= list
;
1636 victim
->list_size
-= sub_size
;
1637 victim
->remaining
-= sub_size
;
1639 target
->list_size
= sub_size
;
1640 target
->remaining
= sub_size
;
1641 target
->working
= 1;
1644 pthread_mutex_lock(&target
->mutex
);
1645 target
->data_ready
= 1;
1646 pthread_cond_signal(&target
->cond
);
1647 pthread_mutex_unlock(&target
->mutex
);
1650 pthread_join(target
->thread
, NULL
);
1651 pthread_cond_destroy(&target
->cond
);
1652 pthread_mutex_destroy(&target
->mutex
);
1659 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1662 static int add_ref_tag(const char *path
, const unsigned char *sha1
, int flag
, void *cb_data
)
1664 unsigned char peeled
[20];
1666 if (!prefixcmp(path
, "refs/tags/") && /* is a tag? */
1667 !peel_ref(path
, peeled
) && /* peelable? */
1668 !is_null_sha1(peeled
) && /* annotated tag? */
1669 locate_object_entry(peeled
)) /* object packed? */
1670 add_object_entry(sha1
, OBJ_TAG
, NULL
, 0);
1674 static void prepare_pack(int window
, int depth
)
1676 struct object_entry
**delta_list
;
1677 uint32_t i
, n
, nr_deltas
;
1679 get_object_details();
1681 if (!nr_objects
|| !window
|| !depth
)
1684 delta_list
= xmalloc(nr_objects
* sizeof(*delta_list
));
1687 for (i
= 0; i
< nr_objects
; i
++) {
1688 struct object_entry
*entry
= objects
+ i
;
1691 /* This happens if we decided to reuse existing
1692 * delta from a pack. "!no_reuse_delta &&" is implied.
1696 if (entry
->size
< 50)
1699 if (entry
->no_try_delta
)
1702 if (!entry
->preferred_base
)
1705 delta_list
[n
++] = entry
;
1708 if (nr_deltas
&& n
> 1) {
1709 unsigned nr_done
= 0;
1711 progress_state
= start_progress("Compressing objects",
1713 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
1714 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
1715 stop_progress(&progress_state
);
1716 if (nr_done
!= nr_deltas
)
1717 die("inconsistency with delta count");
1722 static int git_pack_config(const char *k
, const char *v
)
1724 if(!strcmp(k
, "pack.window")) {
1725 window
= git_config_int(k
, v
);
1728 if (!strcmp(k
, "pack.windowmemory")) {
1729 window_memory_limit
= git_config_ulong(k
, v
);
1732 if (!strcmp(k
, "pack.depth")) {
1733 depth
= git_config_int(k
, v
);
1736 if (!strcmp(k
, "pack.compression")) {
1737 int level
= git_config_int(k
, v
);
1739 level
= Z_DEFAULT_COMPRESSION
;
1740 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
1741 die("bad pack compression level %d", level
);
1742 pack_compression_level
= level
;
1743 pack_compression_seen
= 1;
1746 if (!strcmp(k
, "pack.deltacachesize")) {
1747 max_delta_cache_size
= git_config_int(k
, v
);
1750 if (!strcmp(k
, "pack.deltacachelimit")) {
1751 cache_max_small_delta_size
= git_config_int(k
, v
);
1754 if (!strcmp(k
, "pack.threads")) {
1755 delta_search_threads
= git_config_int(k
, v
);
1756 if (delta_search_threads
< 0)
1757 die("invalid number of threads specified (%d)",
1758 delta_search_threads
);
1759 #ifndef THREADED_DELTA_SEARCH
1760 if (delta_search_threads
!= 1)
1761 warning("no threads support, ignoring %s", k
);
1765 if (!strcmp(k
, "pack.indexversion")) {
1766 pack_idx_default_version
= git_config_int(k
, v
);
1767 if (pack_idx_default_version
> 2)
1768 die("bad pack.indexversion=%d", pack_idx_default_version
);
1771 if (!strcmp(k
, "pack.packsizelimit")) {
1772 pack_size_limit_cfg
= git_config_ulong(k
, v
);
1775 return git_default_config(k
, v
);
1778 static void read_object_list_from_stdin(void)
1780 char line
[40 + 1 + PATH_MAX
+ 2];
1781 unsigned char sha1
[20];
1784 if (!fgets(line
, sizeof(line
), stdin
)) {
1788 die("fgets returned NULL, not EOF, not error!");
1790 die("fgets: %s", strerror(errno
));
1794 if (line
[0] == '-') {
1795 if (get_sha1_hex(line
+1, sha1
))
1796 die("expected edge sha1, got garbage:\n %s",
1798 add_preferred_base(sha1
);
1801 if (get_sha1_hex(line
, sha1
))
1802 die("expected sha1, got garbage:\n %s", line
);
1804 add_preferred_base_object(line
+41);
1805 add_object_entry(sha1
, 0, line
+41, 0);
1809 #define OBJECT_ADDED (1u<<20)
1811 static void show_commit(struct commit
*commit
)
1813 add_object_entry(commit
->object
.sha1
, OBJ_COMMIT
, NULL
, 0);
1814 commit
->object
.flags
|= OBJECT_ADDED
;
1817 static void show_object(struct object_array_entry
*p
)
1819 add_preferred_base_object(p
->name
);
1820 add_object_entry(p
->item
->sha1
, p
->item
->type
, p
->name
, 0);
1821 p
->item
->flags
|= OBJECT_ADDED
;
1824 static void show_edge(struct commit
*commit
)
1826 add_preferred_base(commit
->object
.sha1
);
1829 struct in_pack_object
{
1831 struct object
*object
;
1837 struct in_pack_object
*array
;
1840 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
1842 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->sha1
, p
);
1843 in_pack
->array
[in_pack
->nr
].object
= object
;
1848 * Compare the objects in the offset order, in order to emulate the
1849 * "git-rev-list --objects" output that produced the pack originally.
1851 static int ofscmp(const void *a_
, const void *b_
)
1853 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
1854 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
1856 if (a
->offset
< b
->offset
)
1858 else if (a
->offset
> b
->offset
)
1861 return hashcmp(a
->object
->sha1
, b
->object
->sha1
);
1864 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
1866 struct packed_git
*p
;
1867 struct in_pack in_pack
;
1870 memset(&in_pack
, 0, sizeof(in_pack
));
1872 for (p
= packed_git
; p
; p
= p
->next
) {
1873 const unsigned char *sha1
;
1876 for (i
= 0; i
< revs
->num_ignore_packed
; i
++) {
1877 if (matches_pack_name(p
, revs
->ignore_packed
[i
]))
1880 if (revs
->num_ignore_packed
<= i
)
1882 if (open_pack_index(p
))
1883 die("cannot open pack index");
1885 ALLOC_GROW(in_pack
.array
,
1886 in_pack
.nr
+ p
->num_objects
,
1889 for (i
= 0; i
< p
->num_objects
; i
++) {
1890 sha1
= nth_packed_object_sha1(p
, i
);
1891 o
= lookup_unknown_object(sha1
);
1892 if (!(o
->flags
& OBJECT_ADDED
))
1893 mark_in_pack_object(o
, p
, &in_pack
);
1894 o
->flags
|= OBJECT_ADDED
;
1899 qsort(in_pack
.array
, in_pack
.nr
, sizeof(in_pack
.array
[0]),
1901 for (i
= 0; i
< in_pack
.nr
; i
++) {
1902 struct object
*o
= in_pack
.array
[i
].object
;
1903 add_object_entry(o
->sha1
, o
->type
, "", 0);
1906 free(in_pack
.array
);
1909 static void loosen_unused_packed_objects(struct rev_info
*revs
)
1911 struct packed_git
*p
;
1913 const unsigned char *sha1
;
1915 for (p
= packed_git
; p
; p
= p
->next
) {
1916 for (i
= 0; i
< revs
->num_ignore_packed
; i
++) {
1917 if (matches_pack_name(p
, revs
->ignore_packed
[i
]))
1920 if (revs
->num_ignore_packed
<= i
)
1923 if (open_pack_index(p
))
1924 die("cannot open pack index");
1926 for (i
= 0; i
< p
->num_objects
; i
++) {
1927 sha1
= nth_packed_object_sha1(p
, i
);
1928 if (!locate_object_entry(sha1
))
1929 if (force_object_loose(sha1
, p
->mtime
))
1930 die("unable to force loose object");
1935 static void get_object_list(int ac
, const char **av
)
1937 struct rev_info revs
;
1941 init_revisions(&revs
, NULL
);
1942 save_commit_buffer
= 0;
1943 setup_revisions(ac
, av
, &revs
, NULL
);
1945 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
1946 int len
= strlen(line
);
1947 if (len
&& line
[len
- 1] == '\n')
1952 if (!strcmp(line
, "--not")) {
1953 flags
^= UNINTERESTING
;
1956 die("not a rev '%s'", line
);
1958 if (handle_revision_arg(line
, &revs
, flags
, 1))
1959 die("bad revision '%s'", line
);
1962 if (prepare_revision_walk(&revs
))
1963 die("revision walk setup failed");
1964 mark_edges_uninteresting(revs
.commits
, &revs
, show_edge
);
1965 traverse_commit_list(&revs
, show_commit
, show_object
);
1967 if (keep_unreachable
)
1968 add_objects_in_unpacked_packs(&revs
);
1969 if (unpack_unreachable
)
1970 loosen_unused_packed_objects(&revs
);
1973 static int adjust_perm(const char *path
, mode_t mode
)
1975 if (chmod(path
, mode
))
1977 return adjust_shared_perm(path
);
1980 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
1982 int use_internal_rev_list
= 0;
1986 int rp_ac_alloc
= 64;
1989 rp_av
= xcalloc(rp_ac_alloc
, sizeof(*rp_av
));
1991 rp_av
[0] = "pack-objects";
1992 rp_av
[1] = "--objects"; /* --thin will make it --objects-edge */
1995 git_config(git_pack_config
);
1996 if (!pack_compression_seen
&& core_compression_seen
)
1997 pack_compression_level
= core_compression_level
;
1999 progress
= isatty(2);
2000 for (i
= 1; i
< argc
; i
++) {
2001 const char *arg
= argv
[i
];
2006 if (!strcmp("--non-empty", arg
)) {
2010 if (!strcmp("--local", arg
)) {
2014 if (!strcmp("--incremental", arg
)) {
2018 if (!prefixcmp(arg
, "--compression=")) {
2020 int level
= strtoul(arg
+14, &end
, 0);
2021 if (!arg
[14] || *end
)
2024 level
= Z_DEFAULT_COMPRESSION
;
2025 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
2026 die("bad pack compression level %d", level
);
2027 pack_compression_level
= level
;
2030 if (!prefixcmp(arg
, "--max-pack-size=")) {
2032 pack_size_limit_cfg
= 0;
2033 pack_size_limit
= strtoul(arg
+16, &end
, 0) * 1024 * 1024;
2034 if (!arg
[16] || *end
)
2038 if (!prefixcmp(arg
, "--window=")) {
2040 window
= strtoul(arg
+9, &end
, 0);
2041 if (!arg
[9] || *end
)
2045 if (!prefixcmp(arg
, "--window-memory=")) {
2046 if (!git_parse_ulong(arg
+16, &window_memory_limit
))
2050 if (!prefixcmp(arg
, "--threads=")) {
2052 delta_search_threads
= strtoul(arg
+10, &end
, 0);
2053 if (!arg
[10] || *end
|| delta_search_threads
< 0)
2055 #ifndef THREADED_DELTA_SEARCH
2056 if (delta_search_threads
!= 1)
2057 warning("no threads support, "
2058 "ignoring %s", arg
);
2062 if (!prefixcmp(arg
, "--depth=")) {
2064 depth
= strtoul(arg
+8, &end
, 0);
2065 if (!arg
[8] || *end
)
2069 if (!strcmp("--progress", arg
)) {
2073 if (!strcmp("--all-progress", arg
)) {
2077 if (!strcmp("-q", arg
)) {
2081 if (!strcmp("--no-reuse-delta", arg
)) {
2085 if (!strcmp("--no-reuse-object", arg
)) {
2086 no_reuse_object
= no_reuse_delta
= 1;
2089 if (!strcmp("--delta-base-offset", arg
)) {
2090 allow_ofs_delta
= 1;
2093 if (!strcmp("--stdout", arg
)) {
2097 if (!strcmp("--revs", arg
)) {
2098 use_internal_rev_list
= 1;
2101 if (!strcmp("--keep-unreachable", arg
)) {
2102 keep_unreachable
= 1;
2105 if (!strcmp("--unpack-unreachable", arg
)) {
2106 unpack_unreachable
= 1;
2109 if (!strcmp("--include-tag", arg
)) {
2113 if (!strcmp("--unpacked", arg
) ||
2114 !prefixcmp(arg
, "--unpacked=") ||
2115 !strcmp("--reflog", arg
) ||
2116 !strcmp("--all", arg
)) {
2117 use_internal_rev_list
= 1;
2118 if (rp_ac
>= rp_ac_alloc
- 1) {
2119 rp_ac_alloc
= alloc_nr(rp_ac_alloc
);
2120 rp_av
= xrealloc(rp_av
,
2121 rp_ac_alloc
* sizeof(*rp_av
));
2123 rp_av
[rp_ac
++] = arg
;
2126 if (!strcmp("--thin", arg
)) {
2127 use_internal_rev_list
= 1;
2129 rp_av
[1] = "--objects-edge";
2132 if (!prefixcmp(arg
, "--index-version=")) {
2134 pack_idx_default_version
= strtoul(arg
+ 16, &c
, 10);
2135 if (pack_idx_default_version
> 2)
2138 pack_idx_off32_limit
= strtoul(c
+1, &c
, 0);
2139 if (*c
|| pack_idx_off32_limit
& 0x80000000)
2146 /* Traditionally "pack-objects [options] base extra" failed;
2147 * we would however want to take refs parameter that would
2148 * have been given to upstream rev-list ourselves, which means
2149 * we somehow want to say what the base name is. So the
2152 * pack-objects [options] base <refs...>
2154 * in other words, we would treat the first non-option as the
2155 * base_name and send everything else to the internal revision
2159 if (!pack_to_stdout
)
2160 base_name
= argv
[i
++];
2162 if (pack_to_stdout
!= !base_name
)
2165 if (!pack_to_stdout
&& !pack_size_limit
)
2166 pack_size_limit
= pack_size_limit_cfg
;
2168 if (pack_to_stdout
&& pack_size_limit
)
2169 die("--max-pack-size cannot be used to build a pack for transfer.");
2171 if (!pack_to_stdout
&& thin
)
2172 die("--thin cannot be used to build an indexable pack.");
2174 if (keep_unreachable
&& unpack_unreachable
)
2175 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2177 #ifdef THREADED_DELTA_SEARCH
2178 if (!delta_search_threads
) /* --threads=0 means autodetect */
2179 delta_search_threads
= online_cpus();
2182 prepare_packed_git();
2185 progress_state
= start_progress("Counting objects", 0);
2186 if (!use_internal_rev_list
)
2187 read_object_list_from_stdin();
2189 rp_av
[rp_ac
] = NULL
;
2190 get_object_list(rp_ac
, rp_av
);
2192 if (include_tag
&& nr_result
)
2193 for_each_ref(add_ref_tag
, NULL
);
2194 stop_progress(&progress_state
);
2196 if (non_empty
&& !nr_result
)
2199 prepare_pack(window
, depth
);
2202 fprintf(stderr
, "Total %u (delta %u), reused %u (delta %u)\n",
2203 written
, written_delta
, reused
, reused_delta
);