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] [--keep-unreachable] \n\
32 [<ref-list | <object-list]";
35 struct pack_idx_entry idx
;
36 unsigned long size
; /* uncompressed size */
37 struct packed_git
*in_pack
; /* already in pack */
39 struct object_entry
*delta
; /* delta base object */
40 struct object_entry
*delta_child
; /* deltified objects who bases me */
41 struct object_entry
*delta_sibling
; /* other deltified objects who
42 * uses the same base as me
44 void *delta_data
; /* cached delta (uncompressed) */
45 unsigned long delta_size
; /* delta data size (uncompressed) */
46 unsigned int hash
; /* name hint hash */
47 enum object_type type
;
48 enum object_type in_pack_type
; /* could be delta */
49 unsigned char in_pack_header_size
;
50 unsigned char preferred_base
; /* we do not pack this, but is available
51 * to be used as the base object to delta
54 unsigned char no_try_delta
;
58 * Objects we are going to pack are collected in objects array (dynamically
59 * expanded). nr_objects & nr_alloc controls this array. They are stored
60 * in the order we see -- typically rev-list --objects order that gives us
61 * nice "minimum seek" order.
63 static struct object_entry
*objects
;
64 static struct pack_idx_entry
**written_list
;
65 static uint32_t nr_objects
, nr_alloc
, nr_result
, nr_written
;
68 static int reuse_delta
= 1, reuse_object
= 1;
69 static int keep_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 *get_delta(struct object_entry
*entry
)
108 unsigned long size
, base_size
, delta_size
;
109 void *buf
, *base_buf
, *delta_buf
;
110 enum object_type type
;
112 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
114 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
115 base_buf
= read_sha1_file(entry
->delta
->idx
.sha1
, &type
, &base_size
);
117 die("unable to read %s", sha1_to_hex(entry
->delta
->idx
.sha1
));
118 delta_buf
= diff_delta(base_buf
, base_size
,
119 buf
, size
, &delta_size
, 0);
120 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
121 die("delta size changed");
128 * The per-object header is a pretty dense thing, which is
129 * - first byte: low four bits are "size", then three bits of "type",
130 * and the high bit is "size continues".
131 * - each byte afterwards: low seven bits are size continuation,
132 * with the high bit being "size continues"
134 static int encode_header(enum object_type type
, unsigned long size
, unsigned char *hdr
)
139 if (type
< OBJ_COMMIT
|| type
> OBJ_REF_DELTA
)
140 die("bad type %d", type
);
142 c
= (type
<< 4) | (size
& 15);
155 * we are going to reuse the existing object data as is. make
156 * sure it is not corrupt.
158 static int check_pack_inflate(struct packed_git
*p
,
159 struct pack_window
**w_curs
,
162 unsigned long expect
)
165 unsigned char fakebuf
[4096], *in
;
168 memset(&stream
, 0, sizeof(stream
));
169 inflateInit(&stream
);
171 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
173 stream
.next_out
= fakebuf
;
174 stream
.avail_out
= sizeof(fakebuf
);
175 st
= inflate(&stream
, Z_FINISH
);
176 offset
+= stream
.next_in
- in
;
177 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
179 return (st
== Z_STREAM_END
&&
180 stream
.total_out
== expect
&&
181 stream
.total_in
== len
) ? 0 : -1;
184 static int check_pack_crc(struct packed_git
*p
, struct pack_window
**w_curs
,
185 off_t offset
, off_t len
, unsigned int nr
)
187 const uint32_t *index_crc
;
188 uint32_t data_crc
= crc32(0, Z_NULL
, 0);
192 void *data
= use_pack(p
, w_curs
, offset
, &avail
);
195 data_crc
= crc32(data_crc
, data
, avail
);
200 index_crc
= p
->index_data
;
201 index_crc
+= 2 + 256 + p
->num_objects
* (20/4) + nr
;
203 return data_crc
!= ntohl(*index_crc
);
206 static void copy_pack_data(struct sha1file
*f
,
207 struct packed_git
*p
,
208 struct pack_window
**w_curs
,
216 in
= use_pack(p
, w_curs
, offset
, &avail
);
218 avail
= (unsigned int)len
;
219 sha1write(f
, in
, avail
);
225 static unsigned long write_object(struct sha1file
*f
,
226 struct object_entry
*entry
,
231 unsigned char header
[10];
232 unsigned char dheader
[10];
235 enum object_type obj_type
;
237 /* write limit if limited packsize and not first object */
238 unsigned long limit
= pack_size_limit
&& nr_written
?
239 pack_size_limit
- write_offset
: 0;
241 int usable_delta
= !entry
->delta
? 0 :
242 /* yes if unlimited packfile */
243 !pack_size_limit
? 1 :
244 /* no if base written to previous pack */
245 entry
->delta
->idx
.offset
== (off_t
)-1 ? 0 :
246 /* otherwise double-check written to this
247 * pack, like we do below
249 entry
->delta
->idx
.offset
? 1 : 0;
254 obj_type
= entry
->type
;
256 to_reuse
= 0; /* explicit */
257 else if (!entry
->in_pack
)
258 to_reuse
= 0; /* can't reuse what we don't have */
259 else if (obj_type
== OBJ_REF_DELTA
|| obj_type
== OBJ_OFS_DELTA
)
260 /* check_object() decided it for us ... */
261 to_reuse
= usable_delta
;
262 /* ... but pack split may override that */
263 else if (obj_type
!= entry
->in_pack_type
)
264 to_reuse
= 0; /* pack has delta which is unusable */
265 else if (entry
->delta
)
266 to_reuse
= 0; /* we want to pack afresh */
268 to_reuse
= 1; /* we have it in-pack undeltified,
269 * and we do not need to deltify it.
274 unsigned long maxsize
;
277 buf
= read_sha1_file(entry
->idx
.sha1
, &obj_type
, &size
);
279 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
280 } else if (entry
->delta_data
) {
281 size
= entry
->delta_size
;
282 buf
= entry
->delta_data
;
283 entry
->delta_data
= NULL
;
284 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
285 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
287 buf
= get_delta(entry
);
288 size
= entry
->delta_size
;
289 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
290 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
292 /* compress the data to store and put compressed length in datalen */
293 memset(&stream
, 0, sizeof(stream
));
294 deflateInit(&stream
, pack_compression_level
);
295 maxsize
= deflateBound(&stream
, size
);
296 out
= xmalloc(maxsize
);
298 stream
.next_in
= buf
;
299 stream
.avail_in
= size
;
300 stream
.next_out
= out
;
301 stream
.avail_out
= maxsize
;
302 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
305 datalen
= stream
.total_out
;
308 * The object header is a byte of 'type' followed by zero or
309 * more bytes of length.
311 hdrlen
= encode_header(obj_type
, size
, header
);
313 if (obj_type
== OBJ_OFS_DELTA
) {
315 * Deltas with relative base contain an additional
316 * encoding of the relative offset for the delta
317 * base from this object's position in the pack.
319 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
320 unsigned pos
= sizeof(dheader
) - 1;
321 dheader
[pos
] = ofs
& 127;
323 dheader
[--pos
] = 128 | (--ofs
& 127);
324 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
329 sha1write(f
, header
, hdrlen
);
330 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
331 hdrlen
+= sizeof(dheader
) - pos
;
332 } else if (obj_type
== OBJ_REF_DELTA
) {
334 * Deltas with a base reference contain
335 * an additional 20 bytes for the base sha1.
337 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
342 sha1write(f
, header
, hdrlen
);
343 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
346 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
351 sha1write(f
, header
, hdrlen
);
353 sha1write(f
, out
, datalen
);
358 struct packed_git
*p
= entry
->in_pack
;
359 struct pack_window
*w_curs
= NULL
;
360 struct revindex_entry
*revidx
;
364 obj_type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
365 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
368 hdrlen
= encode_header(obj_type
, entry
->size
, header
);
369 offset
= entry
->in_pack_offset
;
370 revidx
= find_pack_revindex(p
, offset
);
371 datalen
= revidx
[1].offset
- offset
;
372 if (!pack_to_stdout
&& p
->index_version
> 1 &&
373 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
))
374 die("bad packed object CRC for %s", sha1_to_hex(entry
->idx
.sha1
));
375 offset
+= entry
->in_pack_header_size
;
376 datalen
-= entry
->in_pack_header_size
;
377 if (obj_type
== OBJ_OFS_DELTA
) {
378 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
379 unsigned pos
= sizeof(dheader
) - 1;
380 dheader
[pos
] = ofs
& 127;
382 dheader
[--pos
] = 128 | (--ofs
& 127);
383 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
)
385 sha1write(f
, header
, hdrlen
);
386 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
387 hdrlen
+= sizeof(dheader
) - pos
;
388 } else if (obj_type
== OBJ_REF_DELTA
) {
389 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
)
391 sha1write(f
, header
, hdrlen
);
392 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
395 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
)
397 sha1write(f
, header
, hdrlen
);
400 if (!pack_to_stdout
&& p
->index_version
== 1 &&
401 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry
->size
))
402 die("corrupt packed object for %s", sha1_to_hex(entry
->idx
.sha1
));
403 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
411 entry
->idx
.crc32
= crc32_end(f
);
412 return hdrlen
+ datalen
;
415 static off_t
write_one(struct sha1file
*f
,
416 struct object_entry
*e
,
421 /* offset is non zero if object is written already. */
422 if (e
->idx
.offset
|| e
->preferred_base
)
425 /* if we are deltified, write out base object first. */
427 offset
= write_one(f
, e
->delta
, offset
);
432 e
->idx
.offset
= offset
;
433 size
= write_object(f
, e
, offset
);
438 written_list
[nr_written
++] = &e
->idx
;
440 /* make sure off_t is sufficiently large not to wrap */
441 if (offset
> offset
+ size
)
442 die("pack too large for current definition of off_t");
443 return offset
+ size
;
446 /* forward declaration for write_pack_file */
447 static int adjust_perm(const char *path
, mode_t mode
);
449 static void write_pack_file(void)
453 off_t offset
, offset_one
, last_obj_offset
= 0;
454 struct pack_header hdr
;
455 uint32_t nr_remaining
= nr_result
;
456 time_t last_mtime
= 0;
458 if (progress
> pack_to_stdout
)
459 progress_state
= start_progress("Writing objects", nr_result
);
460 written_list
= xmalloc(nr_objects
* sizeof(*written_list
));
463 unsigned char sha1
[20];
464 char *pack_tmp_name
= NULL
;
466 if (pack_to_stdout
) {
467 f
= sha1fd_throughput(1, "<stdout>", progress_state
);
469 char tmpname
[PATH_MAX
];
471 snprintf(tmpname
, sizeof(tmpname
),
472 "%s/tmp_pack_XXXXXX", get_object_directory());
473 fd
= xmkstemp(tmpname
);
474 pack_tmp_name
= xstrdup(tmpname
);
475 f
= sha1fd(fd
, pack_tmp_name
);
478 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
479 hdr
.hdr_version
= htonl(PACK_VERSION
);
480 hdr
.hdr_entries
= htonl(nr_remaining
);
481 sha1write(f
, &hdr
, sizeof(hdr
));
482 offset
= sizeof(hdr
);
484 for (; i
< nr_objects
; i
++) {
485 last_obj_offset
= offset
;
486 offset_one
= write_one(f
, objects
+ i
, offset
);
490 display_progress(progress_state
, written
);
494 * Did we write the wrong # entries in the header?
495 * If so, rewrite it like in fast-import
497 if (pack_to_stdout
|| nr_written
== nr_remaining
) {
498 sha1close(f
, sha1
, 1);
500 int fd
= sha1close(f
, NULL
, 0);
501 fixup_pack_header_footer(fd
, sha1
, pack_tmp_name
, nr_written
);
505 if (!pack_to_stdout
) {
506 mode_t mode
= umask(0);
508 char *idx_tmp_name
, tmpname
[PATH_MAX
];
513 idx_tmp_name
= write_idx_file(NULL
, written_list
,
516 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.pack",
517 base_name
, sha1_to_hex(sha1
));
518 if (adjust_perm(pack_tmp_name
, mode
))
519 die("unable to make temporary pack file readable: %s",
521 if (rename(pack_tmp_name
, tmpname
))
522 die("unable to rename temporary pack file: %s",
526 * Packs are runtime accessed in their mtime
527 * order since newer packs are more likely to contain
528 * younger objects. So if we are creating multiple
529 * packs then we should modify the mtime of later ones
530 * to preserve this property.
532 if (stat(tmpname
, &st
) < 0) {
533 warning("failed to stat %s: %s",
534 tmpname
, strerror(errno
));
535 } else if (!last_mtime
) {
536 last_mtime
= st
.st_mtime
;
539 utb
.actime
= st
.st_atime
;
540 utb
.modtime
= --last_mtime
;
541 if (utime(tmpname
, &utb
) < 0)
542 warning("failed utime() on %s: %s",
543 tmpname
, strerror(errno
));
546 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.idx",
547 base_name
, sha1_to_hex(sha1
));
548 if (adjust_perm(idx_tmp_name
, mode
))
549 die("unable to make temporary index file readable: %s",
551 if (rename(idx_tmp_name
, tmpname
))
552 die("unable to rename temporary index file: %s",
557 puts(sha1_to_hex(sha1
));
560 /* mark written objects as written to previous pack */
561 for (j
= 0; j
< nr_written
; j
++) {
562 written_list
[j
]->offset
= (off_t
)-1;
564 nr_remaining
-= nr_written
;
565 } while (nr_remaining
&& i
< nr_objects
);
568 stop_progress(&progress_state
);
569 if (written
!= nr_result
)
570 die("wrote %u objects while expecting %u", written
, nr_result
);
572 * We have scanned through [0 ... i). Since we have written
573 * the correct number of objects, the remaining [i ... nr_objects)
574 * items must be either already written (due to out-of-order delta base)
575 * or a preferred base. Count those which are neither and complain if any.
577 for (j
= 0; i
< nr_objects
; i
++) {
578 struct object_entry
*e
= objects
+ i
;
579 j
+= !e
->idx
.offset
&& !e
->preferred_base
;
582 die("wrote %u objects as expected but %u unwritten", written
, j
);
585 static int locate_object_entry_hash(const unsigned char *sha1
)
589 memcpy(&ui
, sha1
, sizeof(unsigned int));
590 i
= ui
% object_ix_hashsz
;
591 while (0 < object_ix
[i
]) {
592 if (!hashcmp(sha1
, objects
[object_ix
[i
] - 1].idx
.sha1
))
594 if (++i
== object_ix_hashsz
)
600 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
604 if (!object_ix_hashsz
)
607 i
= locate_object_entry_hash(sha1
);
609 return &objects
[object_ix
[i
]-1];
613 static void rehash_objects(void)
616 struct object_entry
*oe
;
618 object_ix_hashsz
= nr_objects
* 3;
619 if (object_ix_hashsz
< 1024)
620 object_ix_hashsz
= 1024;
621 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
622 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
623 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
624 int ix
= locate_object_entry_hash(oe
->idx
.sha1
);
628 object_ix
[ix
] = i
+ 1;
632 static unsigned name_hash(const char *name
)
641 * This effectively just creates a sortable number from the
642 * last sixteen non-whitespace characters. Last characters
643 * count "most", so things that end in ".c" sort together.
645 while ((c
= *name
++) != 0) {
648 hash
= (hash
>> 2) + (c
<< 24);
653 static void setup_delta_attr_check(struct git_attr_check
*check
)
655 static struct git_attr
*attr_delta
;
658 attr_delta
= git_attr("delta", 5);
660 check
[0].attr
= attr_delta
;
663 static int no_try_delta(const char *path
)
665 struct git_attr_check check
[1];
667 setup_delta_attr_check(check
);
668 if (git_checkattr(path
, ARRAY_SIZE(check
), check
))
670 if (ATTR_FALSE(check
->value
))
675 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
676 const char *name
, int exclude
)
678 struct object_entry
*entry
;
679 struct packed_git
*p
, *found_pack
= NULL
;
680 off_t found_offset
= 0;
682 unsigned hash
= name_hash(name
);
684 ix
= nr_objects
? locate_object_entry_hash(sha1
) : -1;
687 entry
= objects
+ object_ix
[ix
] - 1;
688 if (!entry
->preferred_base
)
690 entry
->preferred_base
= 1;
695 for (p
= packed_git
; p
; p
= p
->next
) {
696 off_t offset
= find_pack_entry_one(sha1
, p
);
699 found_offset
= offset
;
706 if (local
&& !p
->pack_local
)
711 if (nr_objects
>= nr_alloc
) {
712 nr_alloc
= (nr_alloc
+ 1024) * 3 / 2;
713 objects
= xrealloc(objects
, nr_alloc
* sizeof(*entry
));
716 entry
= objects
+ nr_objects
++;
717 memset(entry
, 0, sizeof(*entry
));
718 hashcpy(entry
->idx
.sha1
, sha1
);
723 entry
->preferred_base
= 1;
727 entry
->in_pack
= found_pack
;
728 entry
->in_pack_offset
= found_offset
;
731 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
734 object_ix
[-1 - ix
] = nr_objects
;
736 display_progress(progress_state
, nr_objects
);
738 if (name
&& no_try_delta(name
))
739 entry
->no_try_delta
= 1;
744 struct pbase_tree_cache
{
745 unsigned char sha1
[20];
749 unsigned long tree_size
;
752 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
753 static int pbase_tree_cache_ix(const unsigned char *sha1
)
755 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
757 static int pbase_tree_cache_ix_incr(int ix
)
759 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
762 static struct pbase_tree
{
763 struct pbase_tree
*next
;
764 /* This is a phony "cache" entry; we are not
765 * going to evict it nor find it through _get()
766 * mechanism -- this is for the toplevel node that
767 * would almost always change with any commit.
769 struct pbase_tree_cache pcache
;
772 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
774 struct pbase_tree_cache
*ent
, *nent
;
777 enum object_type type
;
779 int my_ix
= pbase_tree_cache_ix(sha1
);
780 int available_ix
= -1;
782 /* pbase-tree-cache acts as a limited hashtable.
783 * your object will be found at your index or within a few
784 * slots after that slot if it is cached.
786 for (neigh
= 0; neigh
< 8; neigh
++) {
787 ent
= pbase_tree_cache
[my_ix
];
788 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
792 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
793 ((0 <= available_ix
) &&
794 (!ent
&& pbase_tree_cache
[available_ix
])))
795 available_ix
= my_ix
;
798 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
801 /* Did not find one. Either we got a bogus request or
802 * we need to read and perhaps cache.
804 data
= read_sha1_file(sha1
, &type
, &size
);
807 if (type
!= OBJ_TREE
) {
812 /* We need to either cache or return a throwaway copy */
814 if (available_ix
< 0)
817 ent
= pbase_tree_cache
[available_ix
];
818 my_ix
= available_ix
;
822 nent
= xmalloc(sizeof(*nent
));
823 nent
->temporary
= (available_ix
< 0);
826 /* evict and reuse */
827 free(ent
->tree_data
);
830 hashcpy(nent
->sha1
, sha1
);
831 nent
->tree_data
= data
;
832 nent
->tree_size
= size
;
834 if (!nent
->temporary
)
835 pbase_tree_cache
[my_ix
] = nent
;
839 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
841 if (!cache
->temporary
) {
845 free(cache
->tree_data
);
849 static int name_cmp_len(const char *name
)
852 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
857 static void add_pbase_object(struct tree_desc
*tree
,
860 const char *fullname
)
862 struct name_entry entry
;
865 while (tree_entry(tree
,&entry
)) {
866 if (S_ISGITLINK(entry
.mode
))
868 cmp
= tree_entry_len(entry
.path
, entry
.sha1
) != cmplen
? 1 :
869 memcmp(name
, entry
.path
, cmplen
);
874 if (name
[cmplen
] != '/') {
875 add_object_entry(entry
.sha1
,
876 object_type(entry
.mode
),
880 if (S_ISDIR(entry
.mode
)) {
881 struct tree_desc sub
;
882 struct pbase_tree_cache
*tree
;
883 const char *down
= name
+cmplen
+1;
884 int downlen
= name_cmp_len(down
);
886 tree
= pbase_tree_get(entry
.sha1
);
889 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
891 add_pbase_object(&sub
, down
, downlen
, fullname
);
892 pbase_tree_put(tree
);
897 static unsigned *done_pbase_paths
;
898 static int done_pbase_paths_num
;
899 static int done_pbase_paths_alloc
;
900 static int done_pbase_path_pos(unsigned hash
)
903 int hi
= done_pbase_paths_num
;
905 int mi
= (hi
+ lo
) / 2;
906 if (done_pbase_paths
[mi
] == hash
)
908 if (done_pbase_paths
[mi
] < hash
)
916 static int check_pbase_path(unsigned hash
)
918 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
922 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
923 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
924 done_pbase_paths
= xrealloc(done_pbase_paths
,
925 done_pbase_paths_alloc
*
928 done_pbase_paths_num
++;
929 if (pos
< done_pbase_paths_num
)
930 memmove(done_pbase_paths
+ pos
+ 1,
931 done_pbase_paths
+ pos
,
932 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
933 done_pbase_paths
[pos
] = hash
;
937 static void add_preferred_base_object(const char *name
)
939 struct pbase_tree
*it
;
941 unsigned hash
= name_hash(name
);
943 if (!num_preferred_base
|| check_pbase_path(hash
))
946 cmplen
= name_cmp_len(name
);
947 for (it
= pbase_tree
; it
; it
= it
->next
) {
949 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
952 struct tree_desc tree
;
953 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
954 add_pbase_object(&tree
, name
, cmplen
, name
);
959 static void add_preferred_base(unsigned char *sha1
)
961 struct pbase_tree
*it
;
964 unsigned char tree_sha1
[20];
966 if (window
<= num_preferred_base
++)
969 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
973 for (it
= pbase_tree
; it
; it
= it
->next
) {
974 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
980 it
= xcalloc(1, sizeof(*it
));
981 it
->next
= pbase_tree
;
984 hashcpy(it
->pcache
.sha1
, tree_sha1
);
985 it
->pcache
.tree_data
= data
;
986 it
->pcache
.tree_size
= size
;
989 static void check_object(struct object_entry
*entry
)
991 if (entry
->in_pack
) {
992 struct packed_git
*p
= entry
->in_pack
;
993 struct pack_window
*w_curs
= NULL
;
994 const unsigned char *base_ref
= NULL
;
995 struct object_entry
*base_entry
;
996 unsigned long used
, used_0
;
999 unsigned char *buf
, c
;
1001 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1004 * We want in_pack_type even if we do not reuse delta
1005 * since non-delta representations could still be reused.
1007 used
= unpack_object_header_gently(buf
, avail
,
1008 &entry
->in_pack_type
,
1012 * Determine if this is a delta and if so whether we can
1013 * reuse it or not. Otherwise let's find out as cheaply as
1014 * possible what the actual type and size for this object is.
1016 switch (entry
->in_pack_type
) {
1018 /* Not a delta hence we've already got all we need. */
1019 entry
->type
= entry
->in_pack_type
;
1020 entry
->in_pack_header_size
= used
;
1021 unuse_pack(&w_curs
);
1024 if (reuse_delta
&& !entry
->preferred_base
)
1025 base_ref
= use_pack(p
, &w_curs
,
1026 entry
->in_pack_offset
+ used
, NULL
);
1027 entry
->in_pack_header_size
= used
+ 20;
1030 buf
= use_pack(p
, &w_curs
,
1031 entry
->in_pack_offset
+ used
, NULL
);
1037 if (!ofs
|| MSB(ofs
, 7))
1038 die("delta base offset overflow in pack for %s",
1039 sha1_to_hex(entry
->idx
.sha1
));
1041 ofs
= (ofs
<< 7) + (c
& 127);
1043 if (ofs
>= entry
->in_pack_offset
)
1044 die("delta base offset out of bound for %s",
1045 sha1_to_hex(entry
->idx
.sha1
));
1046 ofs
= entry
->in_pack_offset
- ofs
;
1047 if (reuse_delta
&& !entry
->preferred_base
) {
1048 struct revindex_entry
*revidx
;
1049 revidx
= find_pack_revindex(p
, ofs
);
1050 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1052 entry
->in_pack_header_size
= used
+ used_0
;
1056 if (base_ref
&& (base_entry
= locate_object_entry(base_ref
))) {
1058 * If base_ref was set above that means we wish to
1059 * reuse delta data, and we even found that base
1060 * in the list of objects we want to pack. Goodie!
1062 * Depth value does not matter - find_deltas() will
1063 * never consider reused delta as the base object to
1064 * deltify other objects against, in order to avoid
1067 entry
->type
= entry
->in_pack_type
;
1068 entry
->delta
= base_entry
;
1069 entry
->delta_sibling
= base_entry
->delta_child
;
1070 base_entry
->delta_child
= entry
;
1071 unuse_pack(&w_curs
);
1077 * This must be a delta and we already know what the
1078 * final object type is. Let's extract the actual
1079 * object size from the delta header.
1081 entry
->size
= get_size_from_delta(p
, &w_curs
,
1082 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1083 unuse_pack(&w_curs
);
1088 * No choice but to fall back to the recursive delta walk
1089 * with sha1_object_info() to find about the object type
1092 unuse_pack(&w_curs
);
1095 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1096 if (entry
->type
< 0)
1097 die("unable to get type of object %s",
1098 sha1_to_hex(entry
->idx
.sha1
));
1101 static int pack_offset_sort(const void *_a
, const void *_b
)
1103 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1104 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1106 /* avoid filesystem trashing with loose objects */
1107 if (!a
->in_pack
&& !b
->in_pack
)
1108 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1110 if (a
->in_pack
< b
->in_pack
)
1112 if (a
->in_pack
> b
->in_pack
)
1114 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1115 (a
->in_pack_offset
> b
->in_pack_offset
);
1118 static void get_object_details(void)
1121 struct object_entry
**sorted_by_offset
;
1123 sorted_by_offset
= xcalloc(nr_objects
, sizeof(struct object_entry
*));
1124 for (i
= 0; i
< nr_objects
; i
++)
1125 sorted_by_offset
[i
] = objects
+ i
;
1126 qsort(sorted_by_offset
, nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1128 init_pack_revindex();
1130 for (i
= 0; i
< nr_objects
; i
++)
1131 check_object(sorted_by_offset
[i
]);
1133 free(sorted_by_offset
);
1137 * We search for deltas in a list sorted by type, by filename hash, and then
1138 * by size, so that we see progressively smaller and smaller files.
1139 * That's because we prefer deltas to be from the bigger file
1140 * to the smaller -- deletes are potentially cheaper, but perhaps
1141 * more importantly, the bigger file is likely the more recent
1142 * one. The deepest deltas are therefore the oldest objects which are
1143 * less susceptible to be accessed often.
1145 static int type_size_sort(const void *_a
, const void *_b
)
1147 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1148 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1150 if (a
->type
> b
->type
)
1152 if (a
->type
< b
->type
)
1154 if (a
->hash
> b
->hash
)
1156 if (a
->hash
< b
->hash
)
1158 if (a
->preferred_base
> b
->preferred_base
)
1160 if (a
->preferred_base
< b
->preferred_base
)
1162 if (a
->size
> b
->size
)
1164 if (a
->size
< b
->size
)
1166 return a
< b
? -1 : (a
> b
); /* newest first */
1170 struct object_entry
*entry
;
1172 struct delta_index
*index
;
1176 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1177 unsigned long delta_size
)
1179 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1182 if (delta_size
< cache_max_small_delta_size
)
1185 /* cache delta, if objects are large enough compared to delta size */
1186 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1192 #ifdef THREADED_DELTA_SEARCH
1194 static pthread_mutex_t read_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1195 #define read_lock() pthread_mutex_lock(&read_mutex)
1196 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1198 static pthread_mutex_t cache_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1199 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1200 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1202 static pthread_mutex_t progress_mutex
= PTHREAD_MUTEX_INITIALIZER
;
1203 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1204 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1208 #define read_lock() (void)0
1209 #define read_unlock() (void)0
1210 #define cache_lock() (void)0
1211 #define cache_unlock() (void)0
1212 #define progress_lock() (void)0
1213 #define progress_unlock() (void)0
1217 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1218 unsigned max_depth
, unsigned long *mem_usage
)
1220 struct object_entry
*trg_entry
= trg
->entry
;
1221 struct object_entry
*src_entry
= src
->entry
;
1222 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1224 enum object_type type
;
1227 /* Don't bother doing diffs between different types */
1228 if (trg_entry
->type
!= src_entry
->type
)
1232 * We do not bother to try a delta that we discarded
1233 * on an earlier try, but only when reusing delta data.
1235 if (reuse_delta
&& trg_entry
->in_pack
&&
1236 trg_entry
->in_pack
== src_entry
->in_pack
&&
1237 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1238 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1241 /* Let's not bust the allowed depth. */
1242 if (src
->depth
>= max_depth
)
1245 /* Now some size filtering heuristics. */
1246 trg_size
= trg_entry
->size
;
1247 if (!trg_entry
->delta
) {
1248 max_size
= trg_size
/2 - 20;
1251 max_size
= trg_entry
->delta_size
;
1252 ref_depth
= trg
->depth
;
1254 max_size
= max_size
* (max_depth
- src
->depth
) /
1255 (max_depth
- ref_depth
+ 1);
1258 src_size
= src_entry
->size
;
1259 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1260 if (sizediff
>= max_size
)
1262 if (trg_size
< src_size
/ 32)
1265 /* Load data if not already done */
1268 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1271 die("object %s cannot be read",
1272 sha1_to_hex(trg_entry
->idx
.sha1
));
1274 die("object %s inconsistent object length (%lu vs %lu)",
1275 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1280 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1283 die("object %s cannot be read",
1284 sha1_to_hex(src_entry
->idx
.sha1
));
1286 die("object %s inconsistent object length (%lu vs %lu)",
1287 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1291 src
->index
= create_delta_index(src
->data
, src_size
);
1293 static int warned
= 0;
1295 warning("suboptimal pack - out of memory");
1298 *mem_usage
+= sizeof_delta_index(src
->index
);
1301 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1305 if (trg_entry
->delta
) {
1306 /* Prefer only shallower same-sized deltas. */
1307 if (delta_size
== trg_entry
->delta_size
&&
1308 src
->depth
+ 1 >= trg
->depth
) {
1315 * Handle memory allocation outside of the cache
1316 * accounting lock. Compiler will optimize the strangeness
1317 * away when THREADED_DELTA_SEARCH is not defined.
1319 free(trg_entry
->delta_data
);
1321 if (trg_entry
->delta_data
) {
1322 delta_cache_size
-= trg_entry
->delta_size
;
1323 trg_entry
->delta_data
= NULL
;
1325 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1326 delta_cache_size
+= delta_size
;
1328 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1334 trg_entry
->delta
= src_entry
;
1335 trg_entry
->delta_size
= delta_size
;
1336 trg
->depth
= src
->depth
+ 1;
1341 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1343 struct object_entry
*child
= me
->delta_child
;
1346 unsigned int c
= check_delta_limit(child
, n
+ 1);
1349 child
= child
->delta_sibling
;
1354 static unsigned long free_unpacked(struct unpacked
*n
)
1356 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1357 free_delta_index(n
->index
);
1360 freed_mem
+= n
->entry
->size
;
1369 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1370 int window
, int depth
, unsigned *processed
)
1372 uint32_t i
, idx
= 0, count
= 0;
1373 unsigned int array_size
= window
* sizeof(struct unpacked
);
1374 struct unpacked
*array
;
1375 unsigned long mem_usage
= 0;
1377 array
= xmalloc(array_size
);
1378 memset(array
, 0, array_size
);
1381 struct object_entry
*entry
= *list
++;
1382 struct unpacked
*n
= array
+ idx
;
1383 int j
, max_depth
, best_base
= -1;
1391 if (!entry
->preferred_base
) {
1393 display_progress(progress_state
, *processed
);
1397 mem_usage
-= free_unpacked(n
);
1400 while (window_memory_limit
&&
1401 mem_usage
> window_memory_limit
&&
1403 uint32_t tail
= (idx
+ window
- count
) % window
;
1404 mem_usage
-= free_unpacked(array
+ tail
);
1408 /* We do not compute delta to *create* objects we are not
1411 if (entry
->preferred_base
)
1415 * If the current object is at pack edge, take the depth the
1416 * objects that depend on the current object into account
1417 * otherwise they would become too deep.
1420 if (entry
->delta_child
) {
1421 max_depth
-= check_delta_limit(entry
, 0);
1429 uint32_t other_idx
= idx
+ j
;
1431 if (other_idx
>= window
)
1432 other_idx
-= window
;
1433 m
= array
+ other_idx
;
1436 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1440 best_base
= other_idx
;
1443 /* if we made n a delta, and if n is already at max
1444 * depth, leaving it in the window is pointless. we
1445 * should evict it first.
1447 if (entry
->delta
&& depth
<= n
->depth
)
1451 * Move the best delta base up in the window, after the
1452 * currently deltified object, to keep it longer. It will
1453 * be the first base object to be attempted next.
1456 struct unpacked swap
= array
[best_base
];
1457 int dist
= (window
+ idx
- best_base
) % window
;
1458 int dst
= best_base
;
1460 int src
= (dst
+ 1) % window
;
1461 array
[dst
] = array
[src
];
1469 if (count
+ 1 < window
)
1475 for (i
= 0; i
< window
; ++i
) {
1476 free_delta_index(array
[i
].index
);
1477 free(array
[i
].data
);
1482 #ifdef THREADED_DELTA_SEARCH
1485 * The main thread waits on the condition that (at least) one of the workers
1486 * has stopped working (which is indicated in the .working member of
1487 * struct thread_params).
1488 * When a work thread has completed its work, it sets .working to 0 and
1489 * signals the main thread and waits on the condition that .data_ready
1493 struct thread_params
{
1495 struct object_entry
**list
;
1502 pthread_mutex_t mutex
;
1503 pthread_cond_t cond
;
1504 unsigned *processed
;
1507 static pthread_cond_t progress_cond
= PTHREAD_COND_INITIALIZER
;
1509 static void *threaded_find_deltas(void *arg
)
1511 struct thread_params
*me
= arg
;
1513 while (me
->remaining
) {
1514 find_deltas(me
->list
, &me
->remaining
,
1515 me
->window
, me
->depth
, me
->processed
);
1519 pthread_cond_signal(&progress_cond
);
1523 * We must not set ->data_ready before we wait on the
1524 * condition because the main thread may have set it to 1
1525 * before we get here. In order to be sure that new
1526 * work is available if we see 1 in ->data_ready, it
1527 * was initialized to 0 before this thread was spawned
1528 * and we reset it to 0 right away.
1530 pthread_mutex_lock(&me
->mutex
);
1531 while (!me
->data_ready
)
1532 pthread_cond_wait(&me
->cond
, &me
->mutex
);
1534 pthread_mutex_unlock(&me
->mutex
);
1536 /* leave ->working 1 so that this doesn't get more work assigned */
1540 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
1541 int window
, int depth
, unsigned *processed
)
1543 struct thread_params p
[delta_search_threads
];
1544 int i
, ret
, active_threads
= 0;
1546 if (delta_search_threads
<= 1) {
1547 find_deltas(list
, &list_size
, window
, depth
, processed
);
1551 /* Partition the work amongst work threads. */
1552 for (i
= 0; i
< delta_search_threads
; i
++) {
1553 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
1555 p
[i
].window
= window
;
1557 p
[i
].processed
= processed
;
1559 p
[i
].data_ready
= 0;
1561 /* try to split chunks on "path" boundaries */
1562 while (sub_size
&& sub_size
< list_size
&&
1563 list
[sub_size
]->hash
&&
1564 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
1568 p
[i
].list_size
= sub_size
;
1569 p
[i
].remaining
= sub_size
;
1572 list_size
-= sub_size
;
1575 /* Start work threads. */
1576 for (i
= 0; i
< delta_search_threads
; i
++) {
1577 if (!p
[i
].list_size
)
1579 pthread_mutex_init(&p
[i
].mutex
, NULL
);
1580 pthread_cond_init(&p
[i
].cond
, NULL
);
1581 ret
= pthread_create(&p
[i
].thread
, NULL
,
1582 threaded_find_deltas
, &p
[i
]);
1584 die("unable to create thread: %s", strerror(ret
));
1589 * Now let's wait for work completion. Each time a thread is done
1590 * with its work, we steal half of the remaining work from the
1591 * thread with the largest number of unprocessed objects and give
1592 * it to that newly idle thread. This ensure good load balancing
1593 * until the remaining object list segments are simply too short
1594 * to be worth splitting anymore.
1596 while (active_threads
) {
1597 struct thread_params
*target
= NULL
;
1598 struct thread_params
*victim
= NULL
;
1599 unsigned sub_size
= 0;
1603 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
1608 pthread_cond_wait(&progress_cond
, &progress_mutex
);
1611 for (i
= 0; i
< delta_search_threads
; i
++)
1612 if (p
[i
].remaining
> 2*window
&&
1613 (!victim
|| victim
->remaining
< p
[i
].remaining
))
1616 sub_size
= victim
->remaining
/ 2;
1617 list
= victim
->list
+ victim
->list_size
- sub_size
;
1618 while (sub_size
&& list
[0]->hash
&&
1619 list
[0]->hash
== list
[-1]->hash
) {
1625 * It is possible for some "paths" to have
1626 * so many objects that no hash boundary
1627 * might be found. Let's just steal the
1628 * exact half in that case.
1630 sub_size
= victim
->remaining
/ 2;
1633 target
->list
= list
;
1634 victim
->list_size
-= sub_size
;
1635 victim
->remaining
-= sub_size
;
1637 target
->list_size
= sub_size
;
1638 target
->remaining
= sub_size
;
1639 target
->working
= 1;
1642 pthread_mutex_lock(&target
->mutex
);
1643 target
->data_ready
= 1;
1644 pthread_cond_signal(&target
->cond
);
1645 pthread_mutex_unlock(&target
->mutex
);
1648 pthread_join(target
->thread
, NULL
);
1649 pthread_cond_destroy(&target
->cond
);
1650 pthread_mutex_destroy(&target
->mutex
);
1657 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1660 static int add_ref_tag(const char *path
, const unsigned char *sha1
, int flag
, void *cb_data
)
1662 unsigned char peeled
[20];
1664 if (!prefixcmp(path
, "refs/tags/") && /* is a tag? */
1665 !peel_ref(path
, peeled
) && /* peelable? */
1666 !is_null_sha1(peeled
) && /* annotated tag? */
1667 locate_object_entry(peeled
)) /* object packed? */
1668 add_object_entry(sha1
, OBJ_TAG
, NULL
, 0);
1672 static void prepare_pack(int window
, int depth
)
1674 struct object_entry
**delta_list
;
1675 uint32_t i
, n
, nr_deltas
;
1677 get_object_details();
1679 if (!nr_objects
|| !window
|| !depth
)
1682 delta_list
= xmalloc(nr_objects
* sizeof(*delta_list
));
1685 for (i
= 0; i
< nr_objects
; i
++) {
1686 struct object_entry
*entry
= objects
+ i
;
1689 /* This happens if we decided to reuse existing
1690 * delta from a pack. "reuse_delta &&" is implied.
1694 if (entry
->size
< 50)
1697 if (entry
->no_try_delta
)
1700 if (!entry
->preferred_base
)
1703 delta_list
[n
++] = entry
;
1706 if (nr_deltas
&& n
> 1) {
1707 unsigned nr_done
= 0;
1709 progress_state
= start_progress("Compressing objects",
1711 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
1712 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
1713 stop_progress(&progress_state
);
1714 if (nr_done
!= nr_deltas
)
1715 die("inconsistency with delta count");
1720 static int git_pack_config(const char *k
, const char *v
)
1722 if(!strcmp(k
, "pack.window")) {
1723 window
= git_config_int(k
, v
);
1726 if (!strcmp(k
, "pack.windowmemory")) {
1727 window_memory_limit
= git_config_ulong(k
, v
);
1730 if (!strcmp(k
, "pack.depth")) {
1731 depth
= git_config_int(k
, v
);
1734 if (!strcmp(k
, "pack.compression")) {
1735 int level
= git_config_int(k
, v
);
1737 level
= Z_DEFAULT_COMPRESSION
;
1738 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
1739 die("bad pack compression level %d", level
);
1740 pack_compression_level
= level
;
1741 pack_compression_seen
= 1;
1744 if (!strcmp(k
, "pack.deltacachesize")) {
1745 max_delta_cache_size
= git_config_int(k
, v
);
1748 if (!strcmp(k
, "pack.deltacachelimit")) {
1749 cache_max_small_delta_size
= git_config_int(k
, v
);
1752 if (!strcmp(k
, "pack.threads")) {
1753 delta_search_threads
= git_config_int(k
, v
);
1754 if (delta_search_threads
< 0)
1755 die("invalid number of threads specified (%d)",
1756 delta_search_threads
);
1757 #ifndef THREADED_DELTA_SEARCH
1758 if (delta_search_threads
!= 1)
1759 warning("no threads support, ignoring %s", k
);
1763 if (!strcmp(k
, "pack.indexversion")) {
1764 pack_idx_default_version
= git_config_int(k
, v
);
1765 if (pack_idx_default_version
> 2)
1766 die("bad pack.indexversion=%d", pack_idx_default_version
);
1769 if (!strcmp(k
, "pack.packsizelimit")) {
1770 pack_size_limit_cfg
= git_config_ulong(k
, v
);
1773 return git_default_config(k
, v
);
1776 static void read_object_list_from_stdin(void)
1778 char line
[40 + 1 + PATH_MAX
+ 2];
1779 unsigned char sha1
[20];
1782 if (!fgets(line
, sizeof(line
), stdin
)) {
1786 die("fgets returned NULL, not EOF, not error!");
1788 die("fgets: %s", strerror(errno
));
1792 if (line
[0] == '-') {
1793 if (get_sha1_hex(line
+1, sha1
))
1794 die("expected edge sha1, got garbage:\n %s",
1796 add_preferred_base(sha1
);
1799 if (get_sha1_hex(line
, sha1
))
1800 die("expected sha1, got garbage:\n %s", line
);
1802 add_preferred_base_object(line
+41);
1803 add_object_entry(sha1
, 0, line
+41, 0);
1807 #define OBJECT_ADDED (1u<<20)
1809 static void show_commit(struct commit
*commit
)
1811 add_object_entry(commit
->object
.sha1
, OBJ_COMMIT
, NULL
, 0);
1812 commit
->object
.flags
|= OBJECT_ADDED
;
1815 static void show_object(struct object_array_entry
*p
)
1817 add_preferred_base_object(p
->name
);
1818 add_object_entry(p
->item
->sha1
, p
->item
->type
, p
->name
, 0);
1819 p
->item
->flags
|= OBJECT_ADDED
;
1822 static void show_edge(struct commit
*commit
)
1824 add_preferred_base(commit
->object
.sha1
);
1827 struct in_pack_object
{
1829 struct object
*object
;
1835 struct in_pack_object
*array
;
1838 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
1840 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->sha1
, p
);
1841 in_pack
->array
[in_pack
->nr
].object
= object
;
1846 * Compare the objects in the offset order, in order to emulate the
1847 * "git-rev-list --objects" output that produced the pack originally.
1849 static int ofscmp(const void *a_
, const void *b_
)
1851 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
1852 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
1854 if (a
->offset
< b
->offset
)
1856 else if (a
->offset
> b
->offset
)
1859 return hashcmp(a
->object
->sha1
, b
->object
->sha1
);
1862 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
1864 struct packed_git
*p
;
1865 struct in_pack in_pack
;
1868 memset(&in_pack
, 0, sizeof(in_pack
));
1870 for (p
= packed_git
; p
; p
= p
->next
) {
1871 const unsigned char *sha1
;
1874 for (i
= 0; i
< revs
->num_ignore_packed
; i
++) {
1875 if (matches_pack_name(p
, revs
->ignore_packed
[i
]))
1878 if (revs
->num_ignore_packed
<= i
)
1880 if (open_pack_index(p
))
1881 die("cannot open pack index");
1883 ALLOC_GROW(in_pack
.array
,
1884 in_pack
.nr
+ p
->num_objects
,
1887 for (i
= 0; i
< p
->num_objects
; i
++) {
1888 sha1
= nth_packed_object_sha1(p
, i
);
1889 o
= lookup_unknown_object(sha1
);
1890 if (!(o
->flags
& OBJECT_ADDED
))
1891 mark_in_pack_object(o
, p
, &in_pack
);
1892 o
->flags
|= OBJECT_ADDED
;
1897 qsort(in_pack
.array
, in_pack
.nr
, sizeof(in_pack
.array
[0]),
1899 for (i
= 0; i
< in_pack
.nr
; i
++) {
1900 struct object
*o
= in_pack
.array
[i
].object
;
1901 add_object_entry(o
->sha1
, o
->type
, "", 0);
1904 free(in_pack
.array
);
1907 static void get_object_list(int ac
, const char **av
)
1909 struct rev_info revs
;
1913 init_revisions(&revs
, NULL
);
1914 save_commit_buffer
= 0;
1915 setup_revisions(ac
, av
, &revs
, NULL
);
1917 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
1918 int len
= strlen(line
);
1919 if (len
&& line
[len
- 1] == '\n')
1924 if (!strcmp(line
, "--not")) {
1925 flags
^= UNINTERESTING
;
1928 die("not a rev '%s'", line
);
1930 if (handle_revision_arg(line
, &revs
, flags
, 1))
1931 die("bad revision '%s'", line
);
1934 if (prepare_revision_walk(&revs
))
1935 die("revision walk setup failed");
1936 mark_edges_uninteresting(revs
.commits
, &revs
, show_edge
);
1937 traverse_commit_list(&revs
, show_commit
, show_object
);
1939 if (keep_unreachable
)
1940 add_objects_in_unpacked_packs(&revs
);
1943 static int adjust_perm(const char *path
, mode_t mode
)
1945 if (chmod(path
, mode
))
1947 return adjust_shared_perm(path
);
1950 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
1952 int use_internal_rev_list
= 0;
1956 int rp_ac_alloc
= 64;
1959 rp_av
= xcalloc(rp_ac_alloc
, sizeof(*rp_av
));
1961 rp_av
[0] = "pack-objects";
1962 rp_av
[1] = "--objects"; /* --thin will make it --objects-edge */
1965 git_config(git_pack_config
);
1966 if (!pack_compression_seen
&& core_compression_seen
)
1967 pack_compression_level
= core_compression_level
;
1969 progress
= isatty(2);
1970 for (i
= 1; i
< argc
; i
++) {
1971 const char *arg
= argv
[i
];
1976 if (!strcmp("--non-empty", arg
)) {
1980 if (!strcmp("--local", arg
)) {
1984 if (!strcmp("--incremental", arg
)) {
1988 if (!prefixcmp(arg
, "--compression=")) {
1990 int level
= strtoul(arg
+14, &end
, 0);
1991 if (!arg
[14] || *end
)
1994 level
= Z_DEFAULT_COMPRESSION
;
1995 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
1996 die("bad pack compression level %d", level
);
1997 pack_compression_level
= level
;
2000 if (!prefixcmp(arg
, "--max-pack-size=")) {
2002 pack_size_limit_cfg
= 0;
2003 pack_size_limit
= strtoul(arg
+16, &end
, 0) * 1024 * 1024;
2004 if (!arg
[16] || *end
)
2008 if (!prefixcmp(arg
, "--window=")) {
2010 window
= strtoul(arg
+9, &end
, 0);
2011 if (!arg
[9] || *end
)
2015 if (!prefixcmp(arg
, "--window-memory=")) {
2016 if (!git_parse_ulong(arg
+16, &window_memory_limit
))
2020 if (!prefixcmp(arg
, "--threads=")) {
2022 delta_search_threads
= strtoul(arg
+10, &end
, 0);
2023 if (!arg
[10] || *end
|| delta_search_threads
< 0)
2025 #ifndef THREADED_DELTA_SEARCH
2026 if (delta_search_threads
!= 1)
2027 warning("no threads support, "
2028 "ignoring %s", arg
);
2032 if (!prefixcmp(arg
, "--depth=")) {
2034 depth
= strtoul(arg
+8, &end
, 0);
2035 if (!arg
[8] || *end
)
2039 if (!strcmp("--progress", arg
)) {
2043 if (!strcmp("--all-progress", arg
)) {
2047 if (!strcmp("-q", arg
)) {
2051 if (!strcmp("--no-reuse-delta", arg
)) {
2055 if (!strcmp("--no-reuse-object", arg
)) {
2056 reuse_object
= reuse_delta
= 0;
2059 if (!strcmp("--delta-base-offset", arg
)) {
2060 allow_ofs_delta
= 1;
2063 if (!strcmp("--stdout", arg
)) {
2067 if (!strcmp("--revs", arg
)) {
2068 use_internal_rev_list
= 1;
2071 if (!strcmp("--keep-unreachable", arg
)) {
2072 keep_unreachable
= 1;
2075 if (!strcmp("--include-tag", arg
)) {
2079 if (!strcmp("--unpacked", arg
) ||
2080 !prefixcmp(arg
, "--unpacked=") ||
2081 !strcmp("--reflog", arg
) ||
2082 !strcmp("--all", arg
)) {
2083 use_internal_rev_list
= 1;
2084 if (rp_ac
>= rp_ac_alloc
- 1) {
2085 rp_ac_alloc
= alloc_nr(rp_ac_alloc
);
2086 rp_av
= xrealloc(rp_av
,
2087 rp_ac_alloc
* sizeof(*rp_av
));
2089 rp_av
[rp_ac
++] = arg
;
2092 if (!strcmp("--thin", arg
)) {
2093 use_internal_rev_list
= 1;
2095 rp_av
[1] = "--objects-edge";
2098 if (!prefixcmp(arg
, "--index-version=")) {
2100 pack_idx_default_version
= strtoul(arg
+ 16, &c
, 10);
2101 if (pack_idx_default_version
> 2)
2104 pack_idx_off32_limit
= strtoul(c
+1, &c
, 0);
2105 if (*c
|| pack_idx_off32_limit
& 0x80000000)
2112 /* Traditionally "pack-objects [options] base extra" failed;
2113 * we would however want to take refs parameter that would
2114 * have been given to upstream rev-list ourselves, which means
2115 * we somehow want to say what the base name is. So the
2118 * pack-objects [options] base <refs...>
2120 * in other words, we would treat the first non-option as the
2121 * base_name and send everything else to the internal revision
2125 if (!pack_to_stdout
)
2126 base_name
= argv
[i
++];
2128 if (pack_to_stdout
!= !base_name
)
2131 if (!pack_to_stdout
&& !pack_size_limit
)
2132 pack_size_limit
= pack_size_limit_cfg
;
2134 if (pack_to_stdout
&& pack_size_limit
)
2135 die("--max-pack-size cannot be used to build a pack for transfer.");
2137 if (!pack_to_stdout
&& thin
)
2138 die("--thin cannot be used to build an indexable pack.");
2140 #ifdef THREADED_DELTA_SEARCH
2141 if (!delta_search_threads
) /* --threads=0 means autodetect */
2142 delta_search_threads
= online_cpus();
2145 prepare_packed_git();
2148 progress_state
= start_progress("Counting objects", 0);
2149 if (!use_internal_rev_list
)
2150 read_object_list_from_stdin();
2152 rp_av
[rp_ac
] = NULL
;
2153 get_object_list(rp_ac
, rp_av
);
2155 if (include_tag
&& nr_result
)
2156 for_each_ref(add_ref_tag
, NULL
);
2157 stop_progress(&progress_state
);
2159 if (non_empty
&& !nr_result
)
2162 prepare_pack(window
, depth
);
2165 fprintf(stderr
, "Total %u (delta %u), reused %u (delta %u)\n",
2166 written
, written_delta
, reused
, reused_delta
);