11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
16 #include "list-objects.h"
17 #include "pack-objects.h"
20 #include "streaming.h"
21 #include "thread-utils.h"
22 #include "pack-bitmap.h"
23 #include "reachable.h"
24 #include "sha1-array.h"
25 #include "argv-array.h"
27 static const char *pack_usage
[] = {
28 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
29 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
34 * Objects we are going to pack are collected in the `to_pack` structure.
35 * It contains an array (dynamically expanded) of the object data, and a map
36 * that can resolve SHA1s to their position in the array.
38 static struct packing_data to_pack
;
40 static struct pack_idx_entry
**written_list
;
41 static uint32_t nr_result
, nr_written
;
44 static int reuse_delta
= 1, reuse_object
= 1;
45 static int keep_unreachable
, unpack_unreachable
, include_tag
;
46 static unsigned long unpack_unreachable_expiration
;
47 static int pack_loose_unreachable
;
49 static int have_non_local_packs
;
50 static int incremental
;
51 static int ignore_packed_keep
;
52 static int allow_ofs_delta
;
53 static struct pack_idx_option pack_idx_opts
;
54 static const char *base_name
;
55 static int progress
= 1;
56 static int window
= 10;
57 static unsigned long pack_size_limit
;
58 static int depth
= 50;
59 static int delta_search_threads
;
60 static int pack_to_stdout
;
61 static int num_preferred_base
;
62 static struct progress
*progress_state
;
63 static int pack_compression_level
= Z_DEFAULT_COMPRESSION
;
64 static int pack_compression_seen
;
66 static struct packed_git
*reuse_packfile
;
67 static uint32_t reuse_packfile_objects
;
68 static off_t reuse_packfile_offset
;
70 static int use_bitmap_index
= 1;
71 static int write_bitmap_index
;
72 static uint16_t write_bitmap_options
;
74 static unsigned long delta_cache_size
= 0;
75 static unsigned long max_delta_cache_size
= 256 * 1024 * 1024;
76 static unsigned long cache_max_small_delta_size
= 1000;
78 static unsigned long window_memory_limit
= 0;
83 static uint32_t written
, written_delta
;
84 static uint32_t reused
, reused_delta
;
89 static struct commit
**indexed_commits
;
90 static unsigned int indexed_commits_nr
;
91 static unsigned int indexed_commits_alloc
;
93 static void index_commit_for_bitmap(struct commit
*commit
)
95 if (indexed_commits_nr
>= indexed_commits_alloc
) {
96 indexed_commits_alloc
= (indexed_commits_alloc
+ 32) * 2;
97 REALLOC_ARRAY(indexed_commits
, indexed_commits_alloc
);
100 indexed_commits
[indexed_commits_nr
++] = commit
;
103 static void *get_delta(struct object_entry
*entry
)
105 unsigned long size
, base_size
, delta_size
;
106 void *buf
, *base_buf
, *delta_buf
;
107 enum object_type type
;
109 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
111 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
112 base_buf
= read_sha1_file(entry
->delta
->idx
.sha1
, &type
, &base_size
);
114 die("unable to read %s", sha1_to_hex(entry
->delta
->idx
.sha1
));
115 delta_buf
= diff_delta(base_buf
, base_size
,
116 buf
, size
, &delta_size
, 0);
117 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
118 die("delta size changed");
124 static unsigned long do_compress(void **pptr
, unsigned long size
)
128 unsigned long maxsize
;
130 git_deflate_init(&stream
, pack_compression_level
);
131 maxsize
= git_deflate_bound(&stream
, size
);
134 out
= xmalloc(maxsize
);
138 stream
.avail_in
= size
;
139 stream
.next_out
= out
;
140 stream
.avail_out
= maxsize
;
141 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
143 git_deflate_end(&stream
);
146 return stream
.total_out
;
149 static unsigned long write_large_blob_data(struct git_istream
*st
, struct sha1file
*f
,
150 const unsigned char *sha1
)
153 unsigned char ibuf
[1024 * 16];
154 unsigned char obuf
[1024 * 16];
155 unsigned long olen
= 0;
157 git_deflate_init(&stream
, pack_compression_level
);
162 readlen
= read_istream(st
, ibuf
, sizeof(ibuf
));
164 die(_("unable to read %s"), sha1_to_hex(sha1
));
166 stream
.next_in
= ibuf
;
167 stream
.avail_in
= readlen
;
168 while ((stream
.avail_in
|| readlen
== 0) &&
169 (zret
== Z_OK
|| zret
== Z_BUF_ERROR
)) {
170 stream
.next_out
= obuf
;
171 stream
.avail_out
= sizeof(obuf
);
172 zret
= git_deflate(&stream
, readlen
? 0 : Z_FINISH
);
173 sha1write(f
, obuf
, stream
.next_out
- obuf
);
174 olen
+= stream
.next_out
- obuf
;
177 die(_("deflate error (%d)"), zret
);
179 if (zret
!= Z_STREAM_END
)
180 die(_("deflate error (%d)"), zret
);
184 git_deflate_end(&stream
);
189 * we are going to reuse the existing object data as is. make
190 * sure it is not corrupt.
192 static int check_pack_inflate(struct packed_git
*p
,
193 struct pack_window
**w_curs
,
196 unsigned long expect
)
199 unsigned char fakebuf
[4096], *in
;
202 memset(&stream
, 0, sizeof(stream
));
203 git_inflate_init(&stream
);
205 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
207 stream
.next_out
= fakebuf
;
208 stream
.avail_out
= sizeof(fakebuf
);
209 st
= git_inflate(&stream
, Z_FINISH
);
210 offset
+= stream
.next_in
- in
;
211 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
212 git_inflate_end(&stream
);
213 return (st
== Z_STREAM_END
&&
214 stream
.total_out
== expect
&&
215 stream
.total_in
== len
) ? 0 : -1;
218 static void copy_pack_data(struct sha1file
*f
,
219 struct packed_git
*p
,
220 struct pack_window
**w_curs
,
228 in
= use_pack(p
, w_curs
, offset
, &avail
);
230 avail
= (unsigned long)len
;
231 sha1write(f
, in
, avail
);
237 /* Return 0 if we will bust the pack-size limit */
238 static unsigned long write_no_reuse_object(struct sha1file
*f
, struct object_entry
*entry
,
239 unsigned long limit
, int usable_delta
)
241 unsigned long size
, datalen
;
242 unsigned char header
[10], dheader
[10];
244 enum object_type type
;
246 struct git_istream
*st
= NULL
;
249 if (entry
->type
== OBJ_BLOB
&&
250 entry
->size
> big_file_threshold
&&
251 (st
= open_istream(entry
->idx
.sha1
, &type
, &size
, NULL
)) != NULL
)
254 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
256 die(_("unable to read %s"), sha1_to_hex(entry
->idx
.sha1
));
259 * make sure no cached delta data remains from a
260 * previous attempt before a pack split occurred.
262 free(entry
->delta_data
);
263 entry
->delta_data
= NULL
;
264 entry
->z_delta_size
= 0;
265 } else if (entry
->delta_data
) {
266 size
= entry
->delta_size
;
267 buf
= entry
->delta_data
;
268 entry
->delta_data
= NULL
;
269 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
270 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
272 buf
= get_delta(entry
);
273 size
= entry
->delta_size
;
274 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
275 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
278 if (st
) /* large blob case, just assume we don't compress well */
280 else if (entry
->z_delta_size
)
281 datalen
= entry
->z_delta_size
;
283 datalen
= do_compress(&buf
, size
);
286 * The object header is a byte of 'type' followed by zero or
287 * more bytes of length.
289 hdrlen
= encode_in_pack_object_header(type
, size
, header
);
291 if (type
== OBJ_OFS_DELTA
) {
293 * Deltas with relative base contain an additional
294 * encoding of the relative offset for the delta
295 * base from this object's position in the pack.
297 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
298 unsigned pos
= sizeof(dheader
) - 1;
299 dheader
[pos
] = ofs
& 127;
301 dheader
[--pos
] = 128 | (--ofs
& 127);
302 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
308 sha1write(f
, header
, hdrlen
);
309 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
310 hdrlen
+= sizeof(dheader
) - pos
;
311 } else if (type
== OBJ_REF_DELTA
) {
313 * Deltas with a base reference contain
314 * an additional 20 bytes for the base sha1.
316 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
322 sha1write(f
, header
, hdrlen
);
323 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
326 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
332 sha1write(f
, header
, hdrlen
);
335 datalen
= write_large_blob_data(st
, f
, entry
->idx
.sha1
);
338 sha1write(f
, buf
, datalen
);
342 return hdrlen
+ datalen
;
345 /* Return 0 if we will bust the pack-size limit */
346 static unsigned long write_reuse_object(struct sha1file
*f
, struct object_entry
*entry
,
347 unsigned long limit
, int usable_delta
)
349 struct packed_git
*p
= entry
->in_pack
;
350 struct pack_window
*w_curs
= NULL
;
351 struct revindex_entry
*revidx
;
353 enum object_type type
= entry
->type
;
354 unsigned long datalen
;
355 unsigned char header
[10], dheader
[10];
359 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
360 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
361 hdrlen
= encode_in_pack_object_header(type
, entry
->size
, header
);
363 offset
= entry
->in_pack_offset
;
364 revidx
= find_pack_revindex(p
, offset
);
365 datalen
= revidx
[1].offset
- offset
;
366 if (!pack_to_stdout
&& p
->index_version
> 1 &&
367 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
)) {
368 error("bad packed object CRC for %s", sha1_to_hex(entry
->idx
.sha1
));
370 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
373 offset
+= entry
->in_pack_header_size
;
374 datalen
-= entry
->in_pack_header_size
;
376 if (!pack_to_stdout
&& p
->index_version
== 1 &&
377 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry
->size
)) {
378 error("corrupt packed object for %s", sha1_to_hex(entry
->idx
.sha1
));
380 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
383 if (type
== OBJ_OFS_DELTA
) {
384 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
385 unsigned pos
= sizeof(dheader
) - 1;
386 dheader
[pos
] = ofs
& 127;
388 dheader
[--pos
] = 128 | (--ofs
& 127);
389 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
393 sha1write(f
, header
, hdrlen
);
394 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
395 hdrlen
+= sizeof(dheader
) - pos
;
397 } else if (type
== OBJ_REF_DELTA
) {
398 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
402 sha1write(f
, header
, hdrlen
);
403 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
407 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
411 sha1write(f
, header
, hdrlen
);
413 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
416 return hdrlen
+ datalen
;
419 /* Return 0 if we will bust the pack-size limit */
420 static unsigned long write_object(struct sha1file
*f
,
421 struct object_entry
*entry
,
424 unsigned long limit
, len
;
425 int usable_delta
, to_reuse
;
430 /* apply size limit if limited packsize and not first object */
431 if (!pack_size_limit
|| !nr_written
)
433 else if (pack_size_limit
<= write_offset
)
435 * the earlier object did not fit the limit; avoid
436 * mistaking this with unlimited (i.e. limit = 0).
440 limit
= pack_size_limit
- write_offset
;
443 usable_delta
= 0; /* no delta */
444 else if (!pack_size_limit
)
445 usable_delta
= 1; /* unlimited packfile */
446 else if (entry
->delta
->idx
.offset
== (off_t
)-1)
447 usable_delta
= 0; /* base was written to another pack */
448 else if (entry
->delta
->idx
.offset
)
449 usable_delta
= 1; /* base already exists in this pack */
451 usable_delta
= 0; /* base could end up in another pack */
454 to_reuse
= 0; /* explicit */
455 else if (!entry
->in_pack
)
456 to_reuse
= 0; /* can't reuse what we don't have */
457 else if (entry
->type
== OBJ_REF_DELTA
|| entry
->type
== OBJ_OFS_DELTA
)
458 /* check_object() decided it for us ... */
459 to_reuse
= usable_delta
;
460 /* ... but pack split may override that */
461 else if (entry
->type
!= entry
->in_pack_type
)
462 to_reuse
= 0; /* pack has delta which is unusable */
463 else if (entry
->delta
)
464 to_reuse
= 0; /* we want to pack afresh */
466 to_reuse
= 1; /* we have it in-pack undeltified,
467 * and we do not need to deltify it.
471 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
473 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
481 entry
->idx
.crc32
= crc32_end(f
);
485 enum write_one_status
{
486 WRITE_ONE_SKIP
= -1, /* already written */
487 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
488 WRITE_ONE_WRITTEN
= 1, /* normal */
489 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
492 static enum write_one_status
write_one(struct sha1file
*f
,
493 struct object_entry
*e
,
500 * we set offset to 1 (which is an impossible value) to mark
501 * the fact that this object is involved in "write its base
502 * first before writing a deltified object" recursion.
504 recursing
= (e
->idx
.offset
== 1);
506 warning("recursive delta detected for object %s",
507 sha1_to_hex(e
->idx
.sha1
));
508 return WRITE_ONE_RECURSIVE
;
509 } else if (e
->idx
.offset
|| e
->preferred_base
) {
510 /* offset is non zero if object is written already. */
511 return WRITE_ONE_SKIP
;
514 /* if we are deltified, write out base object first. */
516 e
->idx
.offset
= 1; /* now recurse */
517 switch (write_one(f
, e
->delta
, offset
)) {
518 case WRITE_ONE_RECURSIVE
:
519 /* we cannot depend on this one */
524 case WRITE_ONE_BREAK
:
525 e
->idx
.offset
= recursing
;
526 return WRITE_ONE_BREAK
;
530 e
->idx
.offset
= *offset
;
531 size
= write_object(f
, e
, *offset
);
533 e
->idx
.offset
= recursing
;
534 return WRITE_ONE_BREAK
;
536 written_list
[nr_written
++] = &e
->idx
;
538 /* make sure off_t is sufficiently large not to wrap */
539 if (signed_add_overflows(*offset
, size
))
540 die("pack too large for current definition of off_t");
542 return WRITE_ONE_WRITTEN
;
545 static int mark_tagged(const char *path
, const struct object_id
*oid
, int flag
,
548 unsigned char peeled
[20];
549 struct object_entry
*entry
= packlist_find(&to_pack
, oid
->hash
, NULL
);
553 if (!peel_ref(path
, peeled
)) {
554 entry
= packlist_find(&to_pack
, peeled
, NULL
);
561 static inline void add_to_write_order(struct object_entry
**wo
,
563 struct object_entry
*e
)
571 static void add_descendants_to_write_order(struct object_entry
**wo
,
573 struct object_entry
*e
)
575 int add_to_order
= 1;
578 struct object_entry
*s
;
579 /* add this node... */
580 add_to_write_order(wo
, endp
, e
);
581 /* all its siblings... */
582 for (s
= e
->delta_sibling
; s
; s
= s
->delta_sibling
) {
583 add_to_write_order(wo
, endp
, s
);
586 /* drop down a level to add left subtree nodes if possible */
587 if (e
->delta_child
) {
592 /* our sibling might have some children, it is next */
593 if (e
->delta_sibling
) {
594 e
= e
->delta_sibling
;
597 /* go back to our parent node */
599 while (e
&& !e
->delta_sibling
) {
600 /* we're on the right side of a subtree, keep
601 * going up until we can go right again */
605 /* done- we hit our original root node */
608 /* pass it off to sibling at this level */
609 e
= e
->delta_sibling
;
614 static void add_family_to_write_order(struct object_entry
**wo
,
616 struct object_entry
*e
)
618 struct object_entry
*root
;
620 for (root
= e
; root
->delta
; root
= root
->delta
)
622 add_descendants_to_write_order(wo
, endp
, root
);
625 static struct object_entry
**compute_write_order(void)
627 unsigned int i
, wo_end
, last_untagged
;
629 struct object_entry
**wo
;
630 struct object_entry
*objects
= to_pack
.objects
;
632 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
633 objects
[i
].tagged
= 0;
634 objects
[i
].filled
= 0;
635 objects
[i
].delta_child
= NULL
;
636 objects
[i
].delta_sibling
= NULL
;
640 * Fully connect delta_child/delta_sibling network.
641 * Make sure delta_sibling is sorted in the original
644 for (i
= to_pack
.nr_objects
; i
> 0;) {
645 struct object_entry
*e
= &objects
[--i
];
648 /* Mark me as the first child */
649 e
->delta_sibling
= e
->delta
->delta_child
;
650 e
->delta
->delta_child
= e
;
654 * Mark objects that are at the tip of tags.
656 for_each_tag_ref(mark_tagged
, NULL
);
659 * Give the objects in the original recency order until
660 * we see a tagged tip.
662 ALLOC_ARRAY(wo
, to_pack
.nr_objects
);
663 for (i
= wo_end
= 0; i
< to_pack
.nr_objects
; i
++) {
664 if (objects
[i
].tagged
)
666 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
671 * Then fill all the tagged tips.
673 for (; i
< to_pack
.nr_objects
; i
++) {
674 if (objects
[i
].tagged
)
675 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
679 * And then all remaining commits and tags.
681 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
682 if (objects
[i
].type
!= OBJ_COMMIT
&&
683 objects
[i
].type
!= OBJ_TAG
)
685 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
689 * And then all the trees.
691 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
692 if (objects
[i
].type
!= OBJ_TREE
)
694 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
698 * Finally all the rest in really tight order
700 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
701 if (!objects
[i
].filled
)
702 add_family_to_write_order(wo
, &wo_end
, &objects
[i
]);
705 if (wo_end
!= to_pack
.nr_objects
)
706 die("ordered %u objects, expected %"PRIu32
, wo_end
, to_pack
.nr_objects
);
711 static off_t
write_reused_pack(struct sha1file
*f
)
713 unsigned char buffer
[8192];
714 off_t to_write
, total
;
717 if (!is_pack_valid(reuse_packfile
))
718 die("packfile is invalid: %s", reuse_packfile
->pack_name
);
720 fd
= git_open_noatime(reuse_packfile
->pack_name
);
722 die_errno("unable to open packfile for reuse: %s",
723 reuse_packfile
->pack_name
);
725 if (lseek(fd
, sizeof(struct pack_header
), SEEK_SET
) == -1)
726 die_errno("unable to seek in reused packfile");
728 if (reuse_packfile_offset
< 0)
729 reuse_packfile_offset
= reuse_packfile
->pack_size
- 20;
731 total
= to_write
= reuse_packfile_offset
- sizeof(struct pack_header
);
734 int read_pack
= xread(fd
, buffer
, sizeof(buffer
));
737 die_errno("unable to read from reused packfile");
739 if (read_pack
> to_write
)
740 read_pack
= to_write
;
742 sha1write(f
, buffer
, read_pack
);
743 to_write
-= read_pack
;
746 * We don't know the actual number of objects written,
747 * only how many bytes written, how many bytes total, and
748 * how many objects total. So we can fake it by pretending all
749 * objects we are writing are the same size. This gives us a
750 * smooth progress meter, and at the end it matches the true
753 written
= reuse_packfile_objects
*
754 (((double)(total
- to_write
)) / total
);
755 display_progress(progress_state
, written
);
759 written
= reuse_packfile_objects
;
760 display_progress(progress_state
, written
);
761 return reuse_packfile_offset
- sizeof(struct pack_header
);
764 static const char no_split_warning
[] = N_(
765 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
768 static void write_pack_file(void)
773 uint32_t nr_remaining
= nr_result
;
774 time_t last_mtime
= 0;
775 struct object_entry
**write_order
;
777 if (progress
> pack_to_stdout
)
778 progress_state
= start_progress(_("Writing objects"), nr_result
);
779 ALLOC_ARRAY(written_list
, to_pack
.nr_objects
);
780 write_order
= compute_write_order();
783 unsigned char sha1
[20];
784 char *pack_tmp_name
= NULL
;
787 f
= sha1fd_throughput(1, "<stdout>", progress_state
);
789 f
= create_tmp_packfile(&pack_tmp_name
);
791 offset
= write_pack_header(f
, nr_remaining
);
793 if (reuse_packfile
) {
795 assert(pack_to_stdout
);
797 packfile_size
= write_reused_pack(f
);
798 offset
+= packfile_size
;
802 for (; i
< to_pack
.nr_objects
; i
++) {
803 struct object_entry
*e
= write_order
[i
];
804 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
806 display_progress(progress_state
, written
);
810 * Did we write the wrong # entries in the header?
811 * If so, rewrite it like in fast-import
813 if (pack_to_stdout
) {
814 sha1close(f
, sha1
, CSUM_CLOSE
);
815 } else if (nr_written
== nr_remaining
) {
816 sha1close(f
, sha1
, CSUM_FSYNC
);
818 int fd
= sha1close(f
, sha1
, 0);
819 fixup_pack_header_footer(fd
, sha1
, pack_tmp_name
,
820 nr_written
, sha1
, offset
);
822 if (write_bitmap_index
) {
823 warning(_(no_split_warning
));
824 write_bitmap_index
= 0;
828 if (!pack_to_stdout
) {
830 struct strbuf tmpname
= STRBUF_INIT
;
833 * Packs are runtime accessed in their mtime
834 * order since newer packs are more likely to contain
835 * younger objects. So if we are creating multiple
836 * packs then we should modify the mtime of later ones
837 * to preserve this property.
839 if (stat(pack_tmp_name
, &st
) < 0) {
840 warning_errno("failed to stat %s", pack_tmp_name
);
841 } else if (!last_mtime
) {
842 last_mtime
= st
.st_mtime
;
845 utb
.actime
= st
.st_atime
;
846 utb
.modtime
= --last_mtime
;
847 if (utime(pack_tmp_name
, &utb
) < 0)
848 warning_errno("failed utime() on %s", pack_tmp_name
);
851 strbuf_addf(&tmpname
, "%s-", base_name
);
853 if (write_bitmap_index
) {
854 bitmap_writer_set_checksum(sha1
);
855 bitmap_writer_build_type_index(written_list
, nr_written
);
858 finish_tmp_packfile(&tmpname
, pack_tmp_name
,
859 written_list
, nr_written
,
860 &pack_idx_opts
, sha1
);
862 if (write_bitmap_index
) {
863 strbuf_addf(&tmpname
, "%s.bitmap", sha1_to_hex(sha1
));
865 stop_progress(&progress_state
);
867 bitmap_writer_show_progress(progress
);
868 bitmap_writer_reuse_bitmaps(&to_pack
);
869 bitmap_writer_select_commits(indexed_commits
, indexed_commits_nr
, -1);
870 bitmap_writer_build(&to_pack
);
871 bitmap_writer_finish(written_list
, nr_written
,
872 tmpname
.buf
, write_bitmap_options
);
873 write_bitmap_index
= 0;
876 strbuf_release(&tmpname
);
878 puts(sha1_to_hex(sha1
));
881 /* mark written objects as written to previous pack */
882 for (j
= 0; j
< nr_written
; j
++) {
883 written_list
[j
]->offset
= (off_t
)-1;
885 nr_remaining
-= nr_written
;
886 } while (nr_remaining
&& i
< to_pack
.nr_objects
);
890 stop_progress(&progress_state
);
891 if (written
!= nr_result
)
892 die("wrote %"PRIu32
" objects while expecting %"PRIu32
,
896 static void setup_delta_attr_check(struct git_attr_check
*check
)
898 static struct git_attr
*attr_delta
;
901 attr_delta
= git_attr("delta");
903 check
[0].attr
= attr_delta
;
906 static int no_try_delta(const char *path
)
908 struct git_attr_check check
[1];
910 setup_delta_attr_check(check
);
911 if (git_check_attr(path
, ARRAY_SIZE(check
), check
))
913 if (ATTR_FALSE(check
->value
))
919 * When adding an object, check whether we have already added it
920 * to our packing list. If so, we can skip. However, if we are
921 * being asked to excludei t, but the previous mention was to include
922 * it, make sure to adjust its flags and tweak our numbers accordingly.
924 * As an optimization, we pass out the index position where we would have
925 * found the item, since that saves us from having to look it up again a
926 * few lines later when we want to add the new entry.
928 static int have_duplicate_entry(const unsigned char *sha1
,
932 struct object_entry
*entry
;
934 entry
= packlist_find(&to_pack
, sha1
, index_pos
);
939 if (!entry
->preferred_base
)
941 entry
->preferred_base
= 1;
947 static int want_found_object(int exclude
, struct packed_git
*p
)
955 * When asked to do --local (do not include an object that appears in a
956 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
957 * an object that appears in a pack marked with .keep), finding a pack
958 * that matches the criteria is sufficient for us to decide to omit it.
959 * However, even if this pack does not satisfy the criteria, we need to
960 * make sure no copy of this object appears in _any_ pack that makes us
961 * to omit the object, so we need to check all the packs.
963 * We can however first check whether these options can possible matter;
964 * if they do not matter we know we want the object in generated pack.
965 * Otherwise, we signal "-1" at the end to tell the caller that we do
966 * not know either way, and it needs to check more packs.
968 if (!ignore_packed_keep
&&
969 (!local
|| !have_non_local_packs
))
972 if (local
&& !p
->pack_local
)
974 if (ignore_packed_keep
&& p
->pack_local
&& p
->pack_keep
)
977 /* we don't know yet; keep looking for more packs */
982 * Check whether we want the object in the pack (e.g., we do not want
983 * objects found in non-local stores if the "--local" option was used).
985 * If the caller already knows an existing pack it wants to take the object
986 * from, that is passed in *found_pack and *found_offset; otherwise this
987 * function finds if there is any pack that has the object and returns the pack
988 * and its offset in these variables.
990 static int want_object_in_pack(const unsigned char *sha1
,
992 struct packed_git
**found_pack
,
995 struct packed_git
*p
;
998 if (!exclude
&& local
&& has_loose_object_nonlocal(sha1
))
1002 * If we already know the pack object lives in, start checks from that
1003 * pack - in the usual case when neither --local was given nor .keep files
1004 * are present we will determine the answer right now.
1007 want
= want_found_object(exclude
, *found_pack
);
1012 for (p
= packed_git
; p
; p
= p
->next
) {
1015 if (p
== *found_pack
)
1016 offset
= *found_offset
;
1018 offset
= find_pack_entry_one(sha1
, p
);
1022 if (!is_pack_valid(p
))
1024 *found_offset
= offset
;
1027 want
= want_found_object(exclude
, p
);
1036 static void create_object_entry(const unsigned char *sha1
,
1037 enum object_type type
,
1042 struct packed_git
*found_pack
,
1045 struct object_entry
*entry
;
1047 entry
= packlist_alloc(&to_pack
, sha1
, index_pos
);
1052 entry
->preferred_base
= 1;
1056 entry
->in_pack
= found_pack
;
1057 entry
->in_pack_offset
= found_offset
;
1060 entry
->no_try_delta
= no_try_delta
;
1063 static const char no_closure_warning
[] = N_(
1064 "disabling bitmap writing, as some objects are not being packed"
1067 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
1068 const char *name
, int exclude
)
1070 struct packed_git
*found_pack
= NULL
;
1071 off_t found_offset
= 0;
1074 if (have_duplicate_entry(sha1
, exclude
, &index_pos
))
1077 if (!want_object_in_pack(sha1
, exclude
, &found_pack
, &found_offset
)) {
1078 /* The pack is missing an object, so it will not have closure */
1079 if (write_bitmap_index
) {
1080 warning(_(no_closure_warning
));
1081 write_bitmap_index
= 0;
1086 create_object_entry(sha1
, type
, pack_name_hash(name
),
1087 exclude
, name
&& no_try_delta(name
),
1088 index_pos
, found_pack
, found_offset
);
1090 display_progress(progress_state
, nr_result
);
1094 static int add_object_entry_from_bitmap(const unsigned char *sha1
,
1095 enum object_type type
,
1096 int flags
, uint32_t name_hash
,
1097 struct packed_git
*pack
, off_t offset
)
1101 if (have_duplicate_entry(sha1
, 0, &index_pos
))
1104 if (!want_object_in_pack(sha1
, 0, &pack
, &offset
))
1107 create_object_entry(sha1
, type
, name_hash
, 0, 0, index_pos
, pack
, offset
);
1109 display_progress(progress_state
, nr_result
);
1113 struct pbase_tree_cache
{
1114 unsigned char sha1
[20];
1118 unsigned long tree_size
;
1121 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1122 static int pbase_tree_cache_ix(const unsigned char *sha1
)
1124 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
1126 static int pbase_tree_cache_ix_incr(int ix
)
1128 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1131 static struct pbase_tree
{
1132 struct pbase_tree
*next
;
1133 /* This is a phony "cache" entry; we are not
1134 * going to evict it or find it through _get()
1135 * mechanism -- this is for the toplevel node that
1136 * would almost always change with any commit.
1138 struct pbase_tree_cache pcache
;
1141 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
1143 struct pbase_tree_cache
*ent
, *nent
;
1146 enum object_type type
;
1148 int my_ix
= pbase_tree_cache_ix(sha1
);
1149 int available_ix
= -1;
1151 /* pbase-tree-cache acts as a limited hashtable.
1152 * your object will be found at your index or within a few
1153 * slots after that slot if it is cached.
1155 for (neigh
= 0; neigh
< 8; neigh
++) {
1156 ent
= pbase_tree_cache
[my_ix
];
1157 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
1161 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1162 ((0 <= available_ix
) &&
1163 (!ent
&& pbase_tree_cache
[available_ix
])))
1164 available_ix
= my_ix
;
1167 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1170 /* Did not find one. Either we got a bogus request or
1171 * we need to read and perhaps cache.
1173 data
= read_sha1_file(sha1
, &type
, &size
);
1176 if (type
!= OBJ_TREE
) {
1181 /* We need to either cache or return a throwaway copy */
1183 if (available_ix
< 0)
1186 ent
= pbase_tree_cache
[available_ix
];
1187 my_ix
= available_ix
;
1191 nent
= xmalloc(sizeof(*nent
));
1192 nent
->temporary
= (available_ix
< 0);
1195 /* evict and reuse */
1196 free(ent
->tree_data
);
1199 hashcpy(nent
->sha1
, sha1
);
1200 nent
->tree_data
= data
;
1201 nent
->tree_size
= size
;
1203 if (!nent
->temporary
)
1204 pbase_tree_cache
[my_ix
] = nent
;
1208 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1210 if (!cache
->temporary
) {
1214 free(cache
->tree_data
);
1218 static int name_cmp_len(const char *name
)
1221 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
1226 static void add_pbase_object(struct tree_desc
*tree
,
1229 const char *fullname
)
1231 struct name_entry entry
;
1234 while (tree_entry(tree
,&entry
)) {
1235 if (S_ISGITLINK(entry
.mode
))
1237 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1238 memcmp(name
, entry
.path
, cmplen
);
1243 if (name
[cmplen
] != '/') {
1244 add_object_entry(entry
.oid
->hash
,
1245 object_type(entry
.mode
),
1249 if (S_ISDIR(entry
.mode
)) {
1250 struct tree_desc sub
;
1251 struct pbase_tree_cache
*tree
;
1252 const char *down
= name
+cmplen
+1;
1253 int downlen
= name_cmp_len(down
);
1255 tree
= pbase_tree_get(entry
.oid
->hash
);
1258 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1260 add_pbase_object(&sub
, down
, downlen
, fullname
);
1261 pbase_tree_put(tree
);
1266 static unsigned *done_pbase_paths
;
1267 static int done_pbase_paths_num
;
1268 static int done_pbase_paths_alloc
;
1269 static int done_pbase_path_pos(unsigned hash
)
1272 int hi
= done_pbase_paths_num
;
1274 int mi
= (hi
+ lo
) / 2;
1275 if (done_pbase_paths
[mi
] == hash
)
1277 if (done_pbase_paths
[mi
] < hash
)
1285 static int check_pbase_path(unsigned hash
)
1287 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
1291 ALLOC_GROW(done_pbase_paths
,
1292 done_pbase_paths_num
+ 1,
1293 done_pbase_paths_alloc
);
1294 done_pbase_paths_num
++;
1295 if (pos
< done_pbase_paths_num
)
1296 memmove(done_pbase_paths
+ pos
+ 1,
1297 done_pbase_paths
+ pos
,
1298 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
1299 done_pbase_paths
[pos
] = hash
;
1303 static void add_preferred_base_object(const char *name
)
1305 struct pbase_tree
*it
;
1307 unsigned hash
= pack_name_hash(name
);
1309 if (!num_preferred_base
|| check_pbase_path(hash
))
1312 cmplen
= name_cmp_len(name
);
1313 for (it
= pbase_tree
; it
; it
= it
->next
) {
1315 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
1318 struct tree_desc tree
;
1319 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1320 add_pbase_object(&tree
, name
, cmplen
, name
);
1325 static void add_preferred_base(unsigned char *sha1
)
1327 struct pbase_tree
*it
;
1330 unsigned char tree_sha1
[20];
1332 if (window
<= num_preferred_base
++)
1335 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
1339 for (it
= pbase_tree
; it
; it
= it
->next
) {
1340 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
1346 it
= xcalloc(1, sizeof(*it
));
1347 it
->next
= pbase_tree
;
1350 hashcpy(it
->pcache
.sha1
, tree_sha1
);
1351 it
->pcache
.tree_data
= data
;
1352 it
->pcache
.tree_size
= size
;
1355 static void cleanup_preferred_base(void)
1357 struct pbase_tree
*it
;
1363 struct pbase_tree
*this = it
;
1365 free(this->pcache
.tree_data
);
1369 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1370 if (!pbase_tree_cache
[i
])
1372 free(pbase_tree_cache
[i
]->tree_data
);
1373 free(pbase_tree_cache
[i
]);
1374 pbase_tree_cache
[i
] = NULL
;
1377 free(done_pbase_paths
);
1378 done_pbase_paths
= NULL
;
1379 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1382 static void check_object(struct object_entry
*entry
)
1384 if (entry
->in_pack
) {
1385 struct packed_git
*p
= entry
->in_pack
;
1386 struct pack_window
*w_curs
= NULL
;
1387 const unsigned char *base_ref
= NULL
;
1388 struct object_entry
*base_entry
;
1389 unsigned long used
, used_0
;
1390 unsigned long avail
;
1392 unsigned char *buf
, c
;
1394 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1397 * We want in_pack_type even if we do not reuse delta
1398 * since non-delta representations could still be reused.
1400 used
= unpack_object_header_buffer(buf
, avail
,
1401 &entry
->in_pack_type
,
1407 * Determine if this is a delta and if so whether we can
1408 * reuse it or not. Otherwise let's find out as cheaply as
1409 * possible what the actual type and size for this object is.
1411 switch (entry
->in_pack_type
) {
1413 /* Not a delta hence we've already got all we need. */
1414 entry
->type
= entry
->in_pack_type
;
1415 entry
->in_pack_header_size
= used
;
1416 if (entry
->type
< OBJ_COMMIT
|| entry
->type
> OBJ_BLOB
)
1418 unuse_pack(&w_curs
);
1421 if (reuse_delta
&& !entry
->preferred_base
)
1422 base_ref
= use_pack(p
, &w_curs
,
1423 entry
->in_pack_offset
+ used
, NULL
);
1424 entry
->in_pack_header_size
= used
+ 20;
1427 buf
= use_pack(p
, &w_curs
,
1428 entry
->in_pack_offset
+ used
, NULL
);
1434 if (!ofs
|| MSB(ofs
, 7)) {
1435 error("delta base offset overflow in pack for %s",
1436 sha1_to_hex(entry
->idx
.sha1
));
1440 ofs
= (ofs
<< 7) + (c
& 127);
1442 ofs
= entry
->in_pack_offset
- ofs
;
1443 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1444 error("delta base offset out of bound for %s",
1445 sha1_to_hex(entry
->idx
.sha1
));
1448 if (reuse_delta
&& !entry
->preferred_base
) {
1449 struct revindex_entry
*revidx
;
1450 revidx
= find_pack_revindex(p
, ofs
);
1453 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1455 entry
->in_pack_header_size
= used
+ used_0
;
1459 if (base_ref
&& (base_entry
= packlist_find(&to_pack
, base_ref
, NULL
))) {
1461 * If base_ref was set above that means we wish to
1462 * reuse delta data, and we even found that base
1463 * in the list of objects we want to pack. Goodie!
1465 * Depth value does not matter - find_deltas() will
1466 * never consider reused delta as the base object to
1467 * deltify other objects against, in order to avoid
1470 entry
->type
= entry
->in_pack_type
;
1471 entry
->delta
= base_entry
;
1472 entry
->delta_size
= entry
->size
;
1473 entry
->delta_sibling
= base_entry
->delta_child
;
1474 base_entry
->delta_child
= entry
;
1475 unuse_pack(&w_curs
);
1481 * This must be a delta and we already know what the
1482 * final object type is. Let's extract the actual
1483 * object size from the delta header.
1485 entry
->size
= get_size_from_delta(p
, &w_curs
,
1486 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1487 if (entry
->size
== 0)
1489 unuse_pack(&w_curs
);
1494 * No choice but to fall back to the recursive delta walk
1495 * with sha1_object_info() to find about the object type
1499 unuse_pack(&w_curs
);
1502 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1504 * The error condition is checked in prepare_pack(). This is
1505 * to permit a missing preferred base object to be ignored
1506 * as a preferred base. Doing so can result in a larger
1507 * pack file, but the transfer will still take place.
1511 static int pack_offset_sort(const void *_a
, const void *_b
)
1513 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1514 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1516 /* avoid filesystem trashing with loose objects */
1517 if (!a
->in_pack
&& !b
->in_pack
)
1518 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1520 if (a
->in_pack
< b
->in_pack
)
1522 if (a
->in_pack
> b
->in_pack
)
1524 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1525 (a
->in_pack_offset
> b
->in_pack_offset
);
1528 static void get_object_details(void)
1531 struct object_entry
**sorted_by_offset
;
1533 sorted_by_offset
= xcalloc(to_pack
.nr_objects
, sizeof(struct object_entry
*));
1534 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1535 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
1536 qsort(sorted_by_offset
, to_pack
.nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1538 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
1539 struct object_entry
*entry
= sorted_by_offset
[i
];
1540 check_object(entry
);
1541 if (big_file_threshold
< entry
->size
)
1542 entry
->no_try_delta
= 1;
1545 free(sorted_by_offset
);
1549 * We search for deltas in a list sorted by type, by filename hash, and then
1550 * by size, so that we see progressively smaller and smaller files.
1551 * That's because we prefer deltas to be from the bigger file
1552 * to the smaller -- deletes are potentially cheaper, but perhaps
1553 * more importantly, the bigger file is likely the more recent
1554 * one. The deepest deltas are therefore the oldest objects which are
1555 * less susceptible to be accessed often.
1557 static int type_size_sort(const void *_a
, const void *_b
)
1559 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1560 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1562 if (a
->type
> b
->type
)
1564 if (a
->type
< b
->type
)
1566 if (a
->hash
> b
->hash
)
1568 if (a
->hash
< b
->hash
)
1570 if (a
->preferred_base
> b
->preferred_base
)
1572 if (a
->preferred_base
< b
->preferred_base
)
1574 if (a
->size
> b
->size
)
1576 if (a
->size
< b
->size
)
1578 return a
< b
? -1 : (a
> b
); /* newest first */
1582 struct object_entry
*entry
;
1584 struct delta_index
*index
;
1588 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1589 unsigned long delta_size
)
1591 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1594 if (delta_size
< cache_max_small_delta_size
)
1597 /* cache delta, if objects are large enough compared to delta size */
1598 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1606 static pthread_mutex_t read_mutex
;
1607 #define read_lock() pthread_mutex_lock(&read_mutex)
1608 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1610 static pthread_mutex_t cache_mutex
;
1611 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1612 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1614 static pthread_mutex_t progress_mutex
;
1615 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1616 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1620 #define read_lock() (void)0
1621 #define read_unlock() (void)0
1622 #define cache_lock() (void)0
1623 #define cache_unlock() (void)0
1624 #define progress_lock() (void)0
1625 #define progress_unlock() (void)0
1629 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1630 unsigned max_depth
, unsigned long *mem_usage
)
1632 struct object_entry
*trg_entry
= trg
->entry
;
1633 struct object_entry
*src_entry
= src
->entry
;
1634 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1636 enum object_type type
;
1639 /* Don't bother doing diffs between different types */
1640 if (trg_entry
->type
!= src_entry
->type
)
1644 * We do not bother to try a delta that we discarded on an
1645 * earlier try, but only when reusing delta data. Note that
1646 * src_entry that is marked as the preferred_base should always
1647 * be considered, as even if we produce a suboptimal delta against
1648 * it, we will still save the transfer cost, as we already know
1649 * the other side has it and we won't send src_entry at all.
1651 if (reuse_delta
&& trg_entry
->in_pack
&&
1652 trg_entry
->in_pack
== src_entry
->in_pack
&&
1653 !src_entry
->preferred_base
&&
1654 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1655 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1658 /* Let's not bust the allowed depth. */
1659 if (src
->depth
>= max_depth
)
1662 /* Now some size filtering heuristics. */
1663 trg_size
= trg_entry
->size
;
1664 if (!trg_entry
->delta
) {
1665 max_size
= trg_size
/2 - 20;
1668 max_size
= trg_entry
->delta_size
;
1669 ref_depth
= trg
->depth
;
1671 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
1672 (max_depth
- ref_depth
+ 1);
1675 src_size
= src_entry
->size
;
1676 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1677 if (sizediff
>= max_size
)
1679 if (trg_size
< src_size
/ 32)
1682 /* Load data if not already done */
1685 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1688 die("object %s cannot be read",
1689 sha1_to_hex(trg_entry
->idx
.sha1
));
1691 die("object %s inconsistent object length (%lu vs %lu)",
1692 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1697 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1700 if (src_entry
->preferred_base
) {
1701 static int warned
= 0;
1703 warning("object %s cannot be read",
1704 sha1_to_hex(src_entry
->idx
.sha1
));
1706 * Those objects are not included in the
1707 * resulting pack. Be resilient and ignore
1708 * them if they can't be read, in case the
1709 * pack could be created nevertheless.
1713 die("object %s cannot be read",
1714 sha1_to_hex(src_entry
->idx
.sha1
));
1717 die("object %s inconsistent object length (%lu vs %lu)",
1718 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1722 src
->index
= create_delta_index(src
->data
, src_size
);
1724 static int warned
= 0;
1726 warning("suboptimal pack - out of memory");
1729 *mem_usage
+= sizeof_delta_index(src
->index
);
1732 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1736 if (trg_entry
->delta
) {
1737 /* Prefer only shallower same-sized deltas. */
1738 if (delta_size
== trg_entry
->delta_size
&&
1739 src
->depth
+ 1 >= trg
->depth
) {
1746 * Handle memory allocation outside of the cache
1747 * accounting lock. Compiler will optimize the strangeness
1748 * away when NO_PTHREADS is defined.
1750 free(trg_entry
->delta_data
);
1752 if (trg_entry
->delta_data
) {
1753 delta_cache_size
-= trg_entry
->delta_size
;
1754 trg_entry
->delta_data
= NULL
;
1756 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1757 delta_cache_size
+= delta_size
;
1759 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1765 trg_entry
->delta
= src_entry
;
1766 trg_entry
->delta_size
= delta_size
;
1767 trg
->depth
= src
->depth
+ 1;
1772 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1774 struct object_entry
*child
= me
->delta_child
;
1777 unsigned int c
= check_delta_limit(child
, n
+ 1);
1780 child
= child
->delta_sibling
;
1785 static unsigned long free_unpacked(struct unpacked
*n
)
1787 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1788 free_delta_index(n
->index
);
1791 freed_mem
+= n
->entry
->size
;
1800 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1801 int window
, int depth
, unsigned *processed
)
1803 uint32_t i
, idx
= 0, count
= 0;
1804 struct unpacked
*array
;
1805 unsigned long mem_usage
= 0;
1807 array
= xcalloc(window
, sizeof(struct unpacked
));
1810 struct object_entry
*entry
;
1811 struct unpacked
*n
= array
+ idx
;
1812 int j
, max_depth
, best_base
= -1;
1821 if (!entry
->preferred_base
) {
1823 display_progress(progress_state
, *processed
);
1827 mem_usage
-= free_unpacked(n
);
1830 while (window_memory_limit
&&
1831 mem_usage
> window_memory_limit
&&
1833 uint32_t tail
= (idx
+ window
- count
) % window
;
1834 mem_usage
-= free_unpacked(array
+ tail
);
1838 /* We do not compute delta to *create* objects we are not
1841 if (entry
->preferred_base
)
1845 * If the current object is at pack edge, take the depth the
1846 * objects that depend on the current object into account
1847 * otherwise they would become too deep.
1850 if (entry
->delta_child
) {
1851 max_depth
-= check_delta_limit(entry
, 0);
1859 uint32_t other_idx
= idx
+ j
;
1861 if (other_idx
>= window
)
1862 other_idx
-= window
;
1863 m
= array
+ other_idx
;
1866 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1870 best_base
= other_idx
;
1874 * If we decided to cache the delta data, then it is best
1875 * to compress it right away. First because we have to do
1876 * it anyway, and doing it here while we're threaded will
1877 * save a lot of time in the non threaded write phase,
1878 * as well as allow for caching more deltas within
1879 * the same cache size limit.
1881 * But only if not writing to stdout, since in that case
1882 * the network is most likely throttling writes anyway,
1883 * and therefore it is best to go to the write phase ASAP
1884 * instead, as we can afford spending more time compressing
1885 * between writes at that moment.
1887 if (entry
->delta_data
&& !pack_to_stdout
) {
1888 entry
->z_delta_size
= do_compress(&entry
->delta_data
,
1891 delta_cache_size
-= entry
->delta_size
;
1892 delta_cache_size
+= entry
->z_delta_size
;
1896 /* if we made n a delta, and if n is already at max
1897 * depth, leaving it in the window is pointless. we
1898 * should evict it first.
1900 if (entry
->delta
&& max_depth
<= n
->depth
)
1904 * Move the best delta base up in the window, after the
1905 * currently deltified object, to keep it longer. It will
1906 * be the first base object to be attempted next.
1909 struct unpacked swap
= array
[best_base
];
1910 int dist
= (window
+ idx
- best_base
) % window
;
1911 int dst
= best_base
;
1913 int src
= (dst
+ 1) % window
;
1914 array
[dst
] = array
[src
];
1922 if (count
+ 1 < window
)
1928 for (i
= 0; i
< window
; ++i
) {
1929 free_delta_index(array
[i
].index
);
1930 free(array
[i
].data
);
1937 static void try_to_free_from_threads(size_t size
)
1940 release_pack_memory(size
);
1944 static try_to_free_t old_try_to_free_routine
;
1947 * The main thread waits on the condition that (at least) one of the workers
1948 * has stopped working (which is indicated in the .working member of
1949 * struct thread_params).
1950 * When a work thread has completed its work, it sets .working to 0 and
1951 * signals the main thread and waits on the condition that .data_ready
1955 struct thread_params
{
1957 struct object_entry
**list
;
1964 pthread_mutex_t mutex
;
1965 pthread_cond_t cond
;
1966 unsigned *processed
;
1969 static pthread_cond_t progress_cond
;
1972 * Mutex and conditional variable can't be statically-initialized on Windows.
1974 static void init_threaded_search(void)
1976 init_recursive_mutex(&read_mutex
);
1977 pthread_mutex_init(&cache_mutex
, NULL
);
1978 pthread_mutex_init(&progress_mutex
, NULL
);
1979 pthread_cond_init(&progress_cond
, NULL
);
1980 old_try_to_free_routine
= set_try_to_free_routine(try_to_free_from_threads
);
1983 static void cleanup_threaded_search(void)
1985 set_try_to_free_routine(old_try_to_free_routine
);
1986 pthread_cond_destroy(&progress_cond
);
1987 pthread_mutex_destroy(&read_mutex
);
1988 pthread_mutex_destroy(&cache_mutex
);
1989 pthread_mutex_destroy(&progress_mutex
);
1992 static void *threaded_find_deltas(void *arg
)
1994 struct thread_params
*me
= arg
;
1996 while (me
->remaining
) {
1997 find_deltas(me
->list
, &me
->remaining
,
1998 me
->window
, me
->depth
, me
->processed
);
2002 pthread_cond_signal(&progress_cond
);
2006 * We must not set ->data_ready before we wait on the
2007 * condition because the main thread may have set it to 1
2008 * before we get here. In order to be sure that new
2009 * work is available if we see 1 in ->data_ready, it
2010 * was initialized to 0 before this thread was spawned
2011 * and we reset it to 0 right away.
2013 pthread_mutex_lock(&me
->mutex
);
2014 while (!me
->data_ready
)
2015 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2017 pthread_mutex_unlock(&me
->mutex
);
2019 /* leave ->working 1 so that this doesn't get more work assigned */
2023 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2024 int window
, int depth
, unsigned *processed
)
2026 struct thread_params
*p
;
2027 int i
, ret
, active_threads
= 0;
2029 init_threaded_search();
2031 if (delta_search_threads
<= 1) {
2032 find_deltas(list
, &list_size
, window
, depth
, processed
);
2033 cleanup_threaded_search();
2036 if (progress
> pack_to_stdout
)
2037 fprintf(stderr
, "Delta compression using up to %d threads.\n",
2038 delta_search_threads
);
2039 p
= xcalloc(delta_search_threads
, sizeof(*p
));
2041 /* Partition the work amongst work threads. */
2042 for (i
= 0; i
< delta_search_threads
; i
++) {
2043 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2045 /* don't use too small segments or no deltas will be found */
2046 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2049 p
[i
].window
= window
;
2051 p
[i
].processed
= processed
;
2053 p
[i
].data_ready
= 0;
2055 /* try to split chunks on "path" boundaries */
2056 while (sub_size
&& sub_size
< list_size
&&
2057 list
[sub_size
]->hash
&&
2058 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2062 p
[i
].list_size
= sub_size
;
2063 p
[i
].remaining
= sub_size
;
2066 list_size
-= sub_size
;
2069 /* Start work threads. */
2070 for (i
= 0; i
< delta_search_threads
; i
++) {
2071 if (!p
[i
].list_size
)
2073 pthread_mutex_init(&p
[i
].mutex
, NULL
);
2074 pthread_cond_init(&p
[i
].cond
, NULL
);
2075 ret
= pthread_create(&p
[i
].thread
, NULL
,
2076 threaded_find_deltas
, &p
[i
]);
2078 die("unable to create thread: %s", strerror(ret
));
2083 * Now let's wait for work completion. Each time a thread is done
2084 * with its work, we steal half of the remaining work from the
2085 * thread with the largest number of unprocessed objects and give
2086 * it to that newly idle thread. This ensure good load balancing
2087 * until the remaining object list segments are simply too short
2088 * to be worth splitting anymore.
2090 while (active_threads
) {
2091 struct thread_params
*target
= NULL
;
2092 struct thread_params
*victim
= NULL
;
2093 unsigned sub_size
= 0;
2097 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
2102 pthread_cond_wait(&progress_cond
, &progress_mutex
);
2105 for (i
= 0; i
< delta_search_threads
; i
++)
2106 if (p
[i
].remaining
> 2*window
&&
2107 (!victim
|| victim
->remaining
< p
[i
].remaining
))
2110 sub_size
= victim
->remaining
/ 2;
2111 list
= victim
->list
+ victim
->list_size
- sub_size
;
2112 while (sub_size
&& list
[0]->hash
&&
2113 list
[0]->hash
== list
[-1]->hash
) {
2119 * It is possible for some "paths" to have
2120 * so many objects that no hash boundary
2121 * might be found. Let's just steal the
2122 * exact half in that case.
2124 sub_size
= victim
->remaining
/ 2;
2127 target
->list
= list
;
2128 victim
->list_size
-= sub_size
;
2129 victim
->remaining
-= sub_size
;
2131 target
->list_size
= sub_size
;
2132 target
->remaining
= sub_size
;
2133 target
->working
= 1;
2136 pthread_mutex_lock(&target
->mutex
);
2137 target
->data_ready
= 1;
2138 pthread_cond_signal(&target
->cond
);
2139 pthread_mutex_unlock(&target
->mutex
);
2142 pthread_join(target
->thread
, NULL
);
2143 pthread_cond_destroy(&target
->cond
);
2144 pthread_mutex_destroy(&target
->mutex
);
2148 cleanup_threaded_search();
2153 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2156 static int add_ref_tag(const char *path
, const struct object_id
*oid
, int flag
, void *cb_data
)
2158 struct object_id peeled
;
2160 if (starts_with(path
, "refs/tags/") && /* is a tag? */
2161 !peel_ref(path
, peeled
.hash
) && /* peelable? */
2162 packlist_find(&to_pack
, peeled
.hash
, NULL
)) /* object packed? */
2163 add_object_entry(oid
->hash
, OBJ_TAG
, NULL
, 0);
2167 static void prepare_pack(int window
, int depth
)
2169 struct object_entry
**delta_list
;
2170 uint32_t i
, nr_deltas
;
2173 get_object_details();
2176 * If we're locally repacking then we need to be doubly careful
2177 * from now on in order to make sure no stealth corruption gets
2178 * propagated to the new pack. Clients receiving streamed packs
2179 * should validate everything they get anyway so no need to incur
2180 * the additional cost here in that case.
2182 if (!pack_to_stdout
)
2183 do_check_packed_object_crc
= 1;
2185 if (!to_pack
.nr_objects
|| !window
|| !depth
)
2188 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
2191 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2192 struct object_entry
*entry
= to_pack
.objects
+ i
;
2195 /* This happens if we decided to reuse existing
2196 * delta from a pack. "reuse_delta &&" is implied.
2200 if (entry
->size
< 50)
2203 if (entry
->no_try_delta
)
2206 if (!entry
->preferred_base
) {
2208 if (entry
->type
< 0)
2209 die("unable to get type of object %s",
2210 sha1_to_hex(entry
->idx
.sha1
));
2212 if (entry
->type
< 0) {
2214 * This object is not found, but we
2215 * don't have to include it anyway.
2221 delta_list
[n
++] = entry
;
2224 if (nr_deltas
&& n
> 1) {
2225 unsigned nr_done
= 0;
2227 progress_state
= start_progress(_("Compressing objects"),
2229 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
2230 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
2231 stop_progress(&progress_state
);
2232 if (nr_done
!= nr_deltas
)
2233 die("inconsistency with delta count");
2238 static int git_pack_config(const char *k
, const char *v
, void *cb
)
2240 if (!strcmp(k
, "pack.window")) {
2241 window
= git_config_int(k
, v
);
2244 if (!strcmp(k
, "pack.windowmemory")) {
2245 window_memory_limit
= git_config_ulong(k
, v
);
2248 if (!strcmp(k
, "pack.depth")) {
2249 depth
= git_config_int(k
, v
);
2252 if (!strcmp(k
, "pack.compression")) {
2253 int level
= git_config_int(k
, v
);
2255 level
= Z_DEFAULT_COMPRESSION
;
2256 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
2257 die("bad pack compression level %d", level
);
2258 pack_compression_level
= level
;
2259 pack_compression_seen
= 1;
2262 if (!strcmp(k
, "pack.deltacachesize")) {
2263 max_delta_cache_size
= git_config_int(k
, v
);
2266 if (!strcmp(k
, "pack.deltacachelimit")) {
2267 cache_max_small_delta_size
= git_config_int(k
, v
);
2270 if (!strcmp(k
, "pack.writebitmaphashcache")) {
2271 if (git_config_bool(k
, v
))
2272 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
2274 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
2276 if (!strcmp(k
, "pack.usebitmaps")) {
2277 use_bitmap_index
= git_config_bool(k
, v
);
2280 if (!strcmp(k
, "pack.threads")) {
2281 delta_search_threads
= git_config_int(k
, v
);
2282 if (delta_search_threads
< 0)
2283 die("invalid number of threads specified (%d)",
2284 delta_search_threads
);
2286 if (delta_search_threads
!= 1)
2287 warning("no threads support, ignoring %s", k
);
2291 if (!strcmp(k
, "pack.indexversion")) {
2292 pack_idx_opts
.version
= git_config_int(k
, v
);
2293 if (pack_idx_opts
.version
> 2)
2294 die("bad pack.indexversion=%"PRIu32
,
2295 pack_idx_opts
.version
);
2298 return git_default_config(k
, v
, cb
);
2301 static void read_object_list_from_stdin(void)
2303 char line
[40 + 1 + PATH_MAX
+ 2];
2304 unsigned char sha1
[20];
2307 if (!fgets(line
, sizeof(line
), stdin
)) {
2311 die("fgets returned NULL, not EOF, not error!");
2317 if (line
[0] == '-') {
2318 if (get_sha1_hex(line
+1, sha1
))
2319 die("expected edge sha1, got garbage:\n %s",
2321 add_preferred_base(sha1
);
2324 if (get_sha1_hex(line
, sha1
))
2325 die("expected sha1, got garbage:\n %s", line
);
2327 add_preferred_base_object(line
+41);
2328 add_object_entry(sha1
, 0, line
+41, 0);
2332 #define OBJECT_ADDED (1u<<20)
2334 static void show_commit(struct commit
*commit
, void *data
)
2336 add_object_entry(commit
->object
.oid
.hash
, OBJ_COMMIT
, NULL
, 0);
2337 commit
->object
.flags
|= OBJECT_ADDED
;
2339 if (write_bitmap_index
)
2340 index_commit_for_bitmap(commit
);
2343 static void show_object(struct object
*obj
, const char *name
, void *data
)
2345 add_preferred_base_object(name
);
2346 add_object_entry(obj
->oid
.hash
, obj
->type
, name
, 0);
2347 obj
->flags
|= OBJECT_ADDED
;
2350 static void show_edge(struct commit
*commit
)
2352 add_preferred_base(commit
->object
.oid
.hash
);
2355 struct in_pack_object
{
2357 struct object
*object
;
2363 struct in_pack_object
*array
;
2366 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
2368 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->oid
.hash
, p
);
2369 in_pack
->array
[in_pack
->nr
].object
= object
;
2374 * Compare the objects in the offset order, in order to emulate the
2375 * "git rev-list --objects" output that produced the pack originally.
2377 static int ofscmp(const void *a_
, const void *b_
)
2379 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
2380 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
2382 if (a
->offset
< b
->offset
)
2384 else if (a
->offset
> b
->offset
)
2387 return oidcmp(&a
->object
->oid
, &b
->object
->oid
);
2390 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
2392 struct packed_git
*p
;
2393 struct in_pack in_pack
;
2396 memset(&in_pack
, 0, sizeof(in_pack
));
2398 for (p
= packed_git
; p
; p
= p
->next
) {
2399 const unsigned char *sha1
;
2402 if (!p
->pack_local
|| p
->pack_keep
)
2404 if (open_pack_index(p
))
2405 die("cannot open pack index");
2407 ALLOC_GROW(in_pack
.array
,
2408 in_pack
.nr
+ p
->num_objects
,
2411 for (i
= 0; i
< p
->num_objects
; i
++) {
2412 sha1
= nth_packed_object_sha1(p
, i
);
2413 o
= lookup_unknown_object(sha1
);
2414 if (!(o
->flags
& OBJECT_ADDED
))
2415 mark_in_pack_object(o
, p
, &in_pack
);
2416 o
->flags
|= OBJECT_ADDED
;
2421 qsort(in_pack
.array
, in_pack
.nr
, sizeof(in_pack
.array
[0]),
2423 for (i
= 0; i
< in_pack
.nr
; i
++) {
2424 struct object
*o
= in_pack
.array
[i
].object
;
2425 add_object_entry(o
->oid
.hash
, o
->type
, "", 0);
2428 free(in_pack
.array
);
2431 static int add_loose_object(const unsigned char *sha1
, const char *path
,
2434 enum object_type type
= sha1_object_info(sha1
, NULL
);
2437 warning("loose object at %s could not be examined", path
);
2441 add_object_entry(sha1
, type
, "", 0);
2446 * We actually don't even have to worry about reachability here.
2447 * add_object_entry will weed out duplicates, so we just add every
2448 * loose object we find.
2450 static void add_unreachable_loose_objects(void)
2452 for_each_loose_file_in_objdir(get_object_directory(),
2457 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1
)
2459 static struct packed_git
*last_found
= (void *)1;
2460 struct packed_git
*p
;
2462 p
= (last_found
!= (void *)1) ? last_found
: packed_git
;
2465 if ((!p
->pack_local
|| p
->pack_keep
) &&
2466 find_pack_entry_one(sha1
, p
)) {
2470 if (p
== last_found
)
2474 if (p
== last_found
)
2481 * Store a list of sha1s that are should not be discarded
2482 * because they are either written too recently, or are
2483 * reachable from another object that was.
2485 * This is filled by get_object_list.
2487 static struct sha1_array recent_objects
;
2489 static int loosened_object_can_be_discarded(const unsigned char *sha1
,
2490 unsigned long mtime
)
2492 if (!unpack_unreachable_expiration
)
2494 if (mtime
> unpack_unreachable_expiration
)
2496 if (sha1_array_lookup(&recent_objects
, sha1
) >= 0)
2501 static void loosen_unused_packed_objects(struct rev_info
*revs
)
2503 struct packed_git
*p
;
2505 const unsigned char *sha1
;
2507 for (p
= packed_git
; p
; p
= p
->next
) {
2508 if (!p
->pack_local
|| p
->pack_keep
)
2511 if (open_pack_index(p
))
2512 die("cannot open pack index");
2514 for (i
= 0; i
< p
->num_objects
; i
++) {
2515 sha1
= nth_packed_object_sha1(p
, i
);
2516 if (!packlist_find(&to_pack
, sha1
, NULL
) &&
2517 !has_sha1_pack_kept_or_nonlocal(sha1
) &&
2518 !loosened_object_can_be_discarded(sha1
, p
->mtime
))
2519 if (force_object_loose(sha1
, p
->mtime
))
2520 die("unable to force loose object");
2526 * This tracks any options which a reader of the pack might
2527 * not understand, and which would therefore prevent blind reuse
2528 * of what we have on disk.
2530 static int pack_options_allow_reuse(void)
2532 return allow_ofs_delta
;
2535 static int get_object_list_from_bitmap(struct rev_info
*revs
)
2537 if (prepare_bitmap_walk(revs
) < 0)
2540 if (pack_options_allow_reuse() &&
2541 !reuse_partial_packfile_from_bitmap(
2543 &reuse_packfile_objects
,
2544 &reuse_packfile_offset
)) {
2545 assert(reuse_packfile_objects
);
2546 nr_result
+= reuse_packfile_objects
;
2547 display_progress(progress_state
, nr_result
);
2550 traverse_bitmap_commit_list(&add_object_entry_from_bitmap
);
2554 static void record_recent_object(struct object
*obj
,
2558 sha1_array_append(&recent_objects
, obj
->oid
.hash
);
2561 static void record_recent_commit(struct commit
*commit
, void *data
)
2563 sha1_array_append(&recent_objects
, commit
->object
.oid
.hash
);
2566 static void get_object_list(int ac
, const char **av
)
2568 struct rev_info revs
;
2572 init_revisions(&revs
, NULL
);
2573 save_commit_buffer
= 0;
2574 setup_revisions(ac
, av
, &revs
, NULL
);
2576 /* make sure shallows are read */
2577 is_repository_shallow();
2579 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2580 int len
= strlen(line
);
2581 if (len
&& line
[len
- 1] == '\n')
2586 if (!strcmp(line
, "--not")) {
2587 flags
^= UNINTERESTING
;
2588 write_bitmap_index
= 0;
2591 if (starts_with(line
, "--shallow ")) {
2592 unsigned char sha1
[20];
2593 if (get_sha1_hex(line
+ 10, sha1
))
2594 die("not an SHA-1 '%s'", line
+ 10);
2595 register_shallow(sha1
);
2596 use_bitmap_index
= 0;
2599 die("not a rev '%s'", line
);
2601 if (handle_revision_arg(line
, &revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
2602 die("bad revision '%s'", line
);
2605 if (use_bitmap_index
&& !get_object_list_from_bitmap(&revs
))
2608 if (prepare_revision_walk(&revs
))
2609 die("revision walk setup failed");
2610 mark_edges_uninteresting(&revs
, show_edge
);
2611 traverse_commit_list(&revs
, show_commit
, show_object
, NULL
);
2613 if (unpack_unreachable_expiration
) {
2614 revs
.ignore_missing_links
= 1;
2615 if (add_unseen_recent_objects_to_traversal(&revs
,
2616 unpack_unreachable_expiration
))
2617 die("unable to add recent objects");
2618 if (prepare_revision_walk(&revs
))
2619 die("revision walk setup failed");
2620 traverse_commit_list(&revs
, record_recent_commit
,
2621 record_recent_object
, NULL
);
2624 if (keep_unreachable
)
2625 add_objects_in_unpacked_packs(&revs
);
2626 if (pack_loose_unreachable
)
2627 add_unreachable_loose_objects();
2628 if (unpack_unreachable
)
2629 loosen_unused_packed_objects(&revs
);
2631 sha1_array_clear(&recent_objects
);
2634 static int option_parse_index_version(const struct option
*opt
,
2635 const char *arg
, int unset
)
2638 const char *val
= arg
;
2639 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
2640 if (pack_idx_opts
.version
> 2)
2641 die(_("unsupported index version %s"), val
);
2642 if (*c
== ',' && c
[1])
2643 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
2644 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
2645 die(_("bad index version '%s'"), val
);
2649 static int option_parse_unpack_unreachable(const struct option
*opt
,
2650 const char *arg
, int unset
)
2653 unpack_unreachable
= 0;
2654 unpack_unreachable_expiration
= 0;
2657 unpack_unreachable
= 1;
2659 unpack_unreachable_expiration
= approxidate(arg
);
2664 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
2666 int use_internal_rev_list
= 0;
2669 int all_progress_implied
= 0;
2670 struct argv_array rp
= ARGV_ARRAY_INIT
;
2671 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
2672 int rev_list_index
= 0;
2673 struct option pack_objects_options
[] = {
2674 OPT_SET_INT('q', "quiet", &progress
,
2675 N_("do not show progress meter"), 0),
2676 OPT_SET_INT(0, "progress", &progress
,
2677 N_("show progress meter"), 1),
2678 OPT_SET_INT(0, "all-progress", &progress
,
2679 N_("show progress meter during object writing phase"), 2),
2680 OPT_BOOL(0, "all-progress-implied",
2681 &all_progress_implied
,
2682 N_("similar to --all-progress when progress meter is shown")),
2683 { OPTION_CALLBACK
, 0, "index-version", NULL
, N_("version[,offset]"),
2684 N_("write the pack index file in the specified idx format version"),
2685 0, option_parse_index_version
},
2686 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
2687 N_("maximum size of each output pack file")),
2688 OPT_BOOL(0, "local", &local
,
2689 N_("ignore borrowed objects from alternate object store")),
2690 OPT_BOOL(0, "incremental", &incremental
,
2691 N_("ignore packed objects")),
2692 OPT_INTEGER(0, "window", &window
,
2693 N_("limit pack window by objects")),
2694 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
2695 N_("limit pack window by memory in addition to object limit")),
2696 OPT_INTEGER(0, "depth", &depth
,
2697 N_("maximum length of delta chain allowed in the resulting pack")),
2698 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
2699 N_("reuse existing deltas")),
2700 OPT_BOOL(0, "reuse-object", &reuse_object
,
2701 N_("reuse existing objects")),
2702 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
2703 N_("use OFS_DELTA objects")),
2704 OPT_INTEGER(0, "threads", &delta_search_threads
,
2705 N_("use threads when searching for best delta matches")),
2706 OPT_BOOL(0, "non-empty", &non_empty
,
2707 N_("do not create an empty pack output")),
2708 OPT_BOOL(0, "revs", &use_internal_rev_list
,
2709 N_("read revision arguments from standard input")),
2710 { OPTION_SET_INT
, 0, "unpacked", &rev_list_unpacked
, NULL
,
2711 N_("limit the objects to those that are not yet packed"),
2712 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2713 { OPTION_SET_INT
, 0, "all", &rev_list_all
, NULL
,
2714 N_("include objects reachable from any reference"),
2715 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2716 { OPTION_SET_INT
, 0, "reflog", &rev_list_reflog
, NULL
,
2717 N_("include objects referred by reflog entries"),
2718 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2719 { OPTION_SET_INT
, 0, "indexed-objects", &rev_list_index
, NULL
,
2720 N_("include objects referred to by the index"),
2721 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2722 OPT_BOOL(0, "stdout", &pack_to_stdout
,
2723 N_("output pack to stdout")),
2724 OPT_BOOL(0, "include-tag", &include_tag
,
2725 N_("include tag objects that refer to objects to be packed")),
2726 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
2727 N_("keep unreachable objects")),
2728 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
2729 N_("pack loose unreachable objects")),
2730 { OPTION_CALLBACK
, 0, "unpack-unreachable", NULL
, N_("time"),
2731 N_("unpack unreachable objects newer than <time>"),
2732 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
},
2733 OPT_BOOL(0, "thin", &thin
,
2734 N_("create thin packs")),
2735 OPT_BOOL(0, "shallow", &shallow
,
2736 N_("create packs suitable for shallow fetches")),
2737 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep
,
2738 N_("ignore packs that have companion .keep file")),
2739 OPT_INTEGER(0, "compression", &pack_compression_level
,
2740 N_("pack compression level")),
2741 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
2742 N_("do not hide commits by grafts"), 0),
2743 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
2744 N_("use a bitmap index if available to speed up counting objects")),
2745 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index
,
2746 N_("write a bitmap index together with the pack index")),
2750 check_replace_refs
= 0;
2752 reset_pack_idx_option(&pack_idx_opts
);
2753 git_config(git_pack_config
, NULL
);
2754 if (!pack_compression_seen
&& core_compression_seen
)
2755 pack_compression_level
= core_compression_level
;
2757 progress
= isatty(2);
2758 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
2762 base_name
= argv
[0];
2765 if (pack_to_stdout
!= !base_name
|| argc
)
2766 usage_with_options(pack_usage
, pack_objects_options
);
2768 argv_array_push(&rp
, "pack-objects");
2770 use_internal_rev_list
= 1;
2771 argv_array_push(&rp
, shallow
2772 ? "--objects-edge-aggressive"
2773 : "--objects-edge");
2775 argv_array_push(&rp
, "--objects");
2778 use_internal_rev_list
= 1;
2779 argv_array_push(&rp
, "--all");
2781 if (rev_list_reflog
) {
2782 use_internal_rev_list
= 1;
2783 argv_array_push(&rp
, "--reflog");
2785 if (rev_list_index
) {
2786 use_internal_rev_list
= 1;
2787 argv_array_push(&rp
, "--indexed-objects");
2789 if (rev_list_unpacked
) {
2790 use_internal_rev_list
= 1;
2791 argv_array_push(&rp
, "--unpacked");
2796 if (pack_compression_level
== -1)
2797 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
2798 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
2799 die("bad pack compression level %d", pack_compression_level
);
2801 if (!delta_search_threads
) /* --threads=0 means autodetect */
2802 delta_search_threads
= online_cpus();
2805 if (delta_search_threads
!= 1)
2806 warning("no threads support, ignoring --threads");
2808 if (!pack_to_stdout
&& !pack_size_limit
)
2809 pack_size_limit
= pack_size_limit_cfg
;
2810 if (pack_to_stdout
&& pack_size_limit
)
2811 die("--max-pack-size cannot be used to build a pack for transfer.");
2812 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
2813 warning("minimum pack size limit is 1 MiB");
2814 pack_size_limit
= 1024*1024;
2817 if (!pack_to_stdout
&& thin
)
2818 die("--thin cannot be used to build an indexable pack.");
2820 if (keep_unreachable
&& unpack_unreachable
)
2821 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2822 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
2823 unpack_unreachable_expiration
= 0;
2825 if (!use_internal_rev_list
|| !pack_to_stdout
|| is_repository_shallow())
2826 use_bitmap_index
= 0;
2828 if (pack_to_stdout
|| !rev_list_all
)
2829 write_bitmap_index
= 0;
2831 if (progress
&& all_progress_implied
)
2834 prepare_packed_git();
2835 if (ignore_packed_keep
) {
2836 struct packed_git
*p
;
2837 for (p
= packed_git
; p
; p
= p
->next
)
2838 if (p
->pack_local
&& p
->pack_keep
)
2840 if (!p
) /* no keep-able packs found */
2841 ignore_packed_keep
= 0;
2845 * unlike ignore_packed_keep above, we do not want to
2846 * unset "local" based on looking at packs, as it
2847 * also covers non-local objects
2849 struct packed_git
*p
;
2850 for (p
= packed_git
; p
; p
= p
->next
) {
2851 if (!p
->pack_local
) {
2852 have_non_local_packs
= 1;
2859 progress_state
= start_progress(_("Counting objects"), 0);
2860 if (!use_internal_rev_list
)
2861 read_object_list_from_stdin();
2863 get_object_list(rp
.argc
, rp
.argv
);
2864 argv_array_clear(&rp
);
2866 cleanup_preferred_base();
2867 if (include_tag
&& nr_result
)
2868 for_each_ref(add_ref_tag
, NULL
);
2869 stop_progress(&progress_state
);
2871 if (non_empty
&& !nr_result
)
2874 prepare_pack(window
, depth
);
2877 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
2878 " reused %"PRIu32
" (delta %"PRIu32
")\n",
2879 written
, written_delta
, reused
, reused_delta
);