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;
948 * Check whether we want the object in the pack (e.g., we do not want
949 * objects found in non-local stores if the "--local" option was used).
951 * As a side effect of this check, we will find the packed version of this
952 * object, if any. We therefore pass out the pack information to avoid having
953 * to look it up again later.
955 static int want_object_in_pack(const unsigned char *sha1
,
957 struct packed_git
**found_pack
,
960 struct packed_git
*p
;
962 if (!exclude
&& local
&& has_loose_object_nonlocal(sha1
))
968 for (p
= packed_git
; p
; p
= p
->next
) {
969 off_t offset
= find_pack_entry_one(sha1
, p
);
972 if (!is_pack_valid(p
))
974 *found_offset
= offset
;
983 * When asked to do --local (do not include an
984 * object that appears in a pack we borrow
985 * from elsewhere) or --honor-pack-keep (do not
986 * include an object that appears in a pack marked
987 * with .keep), we need to make sure no copy of this
988 * object come from in _any_ pack that causes us to
989 * omit it, and need to complete this loop. When
990 * neither option is in effect, we know the object
991 * we just found is going to be packed, so break
992 * out of the loop to return 1 now.
994 if (!ignore_packed_keep
&&
995 (!local
|| !have_non_local_packs
))
998 if (local
&& !p
->pack_local
)
1000 if (ignore_packed_keep
&& p
->pack_local
&& p
->pack_keep
)
1008 static void create_object_entry(const unsigned char *sha1
,
1009 enum object_type type
,
1014 struct packed_git
*found_pack
,
1017 struct object_entry
*entry
;
1019 entry
= packlist_alloc(&to_pack
, sha1
, index_pos
);
1024 entry
->preferred_base
= 1;
1028 entry
->in_pack
= found_pack
;
1029 entry
->in_pack_offset
= found_offset
;
1032 entry
->no_try_delta
= no_try_delta
;
1035 static const char no_closure_warning
[] = N_(
1036 "disabling bitmap writing, as some objects are not being packed"
1039 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
1040 const char *name
, int exclude
)
1042 struct packed_git
*found_pack
;
1046 if (have_duplicate_entry(sha1
, exclude
, &index_pos
))
1049 if (!want_object_in_pack(sha1
, exclude
, &found_pack
, &found_offset
)) {
1050 /* The pack is missing an object, so it will not have closure */
1051 if (write_bitmap_index
) {
1052 warning(_(no_closure_warning
));
1053 write_bitmap_index
= 0;
1058 create_object_entry(sha1
, type
, pack_name_hash(name
),
1059 exclude
, name
&& no_try_delta(name
),
1060 index_pos
, found_pack
, found_offset
);
1062 display_progress(progress_state
, nr_result
);
1066 static int add_object_entry_from_bitmap(const unsigned char *sha1
,
1067 enum object_type type
,
1068 int flags
, uint32_t name_hash
,
1069 struct packed_git
*pack
, off_t offset
)
1073 if (have_duplicate_entry(sha1
, 0, &index_pos
))
1076 create_object_entry(sha1
, type
, name_hash
, 0, 0, index_pos
, pack
, offset
);
1078 display_progress(progress_state
, nr_result
);
1082 struct pbase_tree_cache
{
1083 unsigned char sha1
[20];
1087 unsigned long tree_size
;
1090 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1091 static int pbase_tree_cache_ix(const unsigned char *sha1
)
1093 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
1095 static int pbase_tree_cache_ix_incr(int ix
)
1097 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1100 static struct pbase_tree
{
1101 struct pbase_tree
*next
;
1102 /* This is a phony "cache" entry; we are not
1103 * going to evict it or find it through _get()
1104 * mechanism -- this is for the toplevel node that
1105 * would almost always change with any commit.
1107 struct pbase_tree_cache pcache
;
1110 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
1112 struct pbase_tree_cache
*ent
, *nent
;
1115 enum object_type type
;
1117 int my_ix
= pbase_tree_cache_ix(sha1
);
1118 int available_ix
= -1;
1120 /* pbase-tree-cache acts as a limited hashtable.
1121 * your object will be found at your index or within a few
1122 * slots after that slot if it is cached.
1124 for (neigh
= 0; neigh
< 8; neigh
++) {
1125 ent
= pbase_tree_cache
[my_ix
];
1126 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
1130 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1131 ((0 <= available_ix
) &&
1132 (!ent
&& pbase_tree_cache
[available_ix
])))
1133 available_ix
= my_ix
;
1136 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1139 /* Did not find one. Either we got a bogus request or
1140 * we need to read and perhaps cache.
1142 data
= read_sha1_file(sha1
, &type
, &size
);
1145 if (type
!= OBJ_TREE
) {
1150 /* We need to either cache or return a throwaway copy */
1152 if (available_ix
< 0)
1155 ent
= pbase_tree_cache
[available_ix
];
1156 my_ix
= available_ix
;
1160 nent
= xmalloc(sizeof(*nent
));
1161 nent
->temporary
= (available_ix
< 0);
1164 /* evict and reuse */
1165 free(ent
->tree_data
);
1168 hashcpy(nent
->sha1
, sha1
);
1169 nent
->tree_data
= data
;
1170 nent
->tree_size
= size
;
1172 if (!nent
->temporary
)
1173 pbase_tree_cache
[my_ix
] = nent
;
1177 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1179 if (!cache
->temporary
) {
1183 free(cache
->tree_data
);
1187 static int name_cmp_len(const char *name
)
1190 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
1195 static void add_pbase_object(struct tree_desc
*tree
,
1198 const char *fullname
)
1200 struct name_entry entry
;
1203 while (tree_entry(tree
,&entry
)) {
1204 if (S_ISGITLINK(entry
.mode
))
1206 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1207 memcmp(name
, entry
.path
, cmplen
);
1212 if (name
[cmplen
] != '/') {
1213 add_object_entry(entry
.oid
->hash
,
1214 object_type(entry
.mode
),
1218 if (S_ISDIR(entry
.mode
)) {
1219 struct tree_desc sub
;
1220 struct pbase_tree_cache
*tree
;
1221 const char *down
= name
+cmplen
+1;
1222 int downlen
= name_cmp_len(down
);
1224 tree
= pbase_tree_get(entry
.oid
->hash
);
1227 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1229 add_pbase_object(&sub
, down
, downlen
, fullname
);
1230 pbase_tree_put(tree
);
1235 static unsigned *done_pbase_paths
;
1236 static int done_pbase_paths_num
;
1237 static int done_pbase_paths_alloc
;
1238 static int done_pbase_path_pos(unsigned hash
)
1241 int hi
= done_pbase_paths_num
;
1243 int mi
= (hi
+ lo
) / 2;
1244 if (done_pbase_paths
[mi
] == hash
)
1246 if (done_pbase_paths
[mi
] < hash
)
1254 static int check_pbase_path(unsigned hash
)
1256 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
1260 ALLOC_GROW(done_pbase_paths
,
1261 done_pbase_paths_num
+ 1,
1262 done_pbase_paths_alloc
);
1263 done_pbase_paths_num
++;
1264 if (pos
< done_pbase_paths_num
)
1265 memmove(done_pbase_paths
+ pos
+ 1,
1266 done_pbase_paths
+ pos
,
1267 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
1268 done_pbase_paths
[pos
] = hash
;
1272 static void add_preferred_base_object(const char *name
)
1274 struct pbase_tree
*it
;
1276 unsigned hash
= pack_name_hash(name
);
1278 if (!num_preferred_base
|| check_pbase_path(hash
))
1281 cmplen
= name_cmp_len(name
);
1282 for (it
= pbase_tree
; it
; it
= it
->next
) {
1284 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
1287 struct tree_desc tree
;
1288 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1289 add_pbase_object(&tree
, name
, cmplen
, name
);
1294 static void add_preferred_base(unsigned char *sha1
)
1296 struct pbase_tree
*it
;
1299 unsigned char tree_sha1
[20];
1301 if (window
<= num_preferred_base
++)
1304 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
1308 for (it
= pbase_tree
; it
; it
= it
->next
) {
1309 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
1315 it
= xcalloc(1, sizeof(*it
));
1316 it
->next
= pbase_tree
;
1319 hashcpy(it
->pcache
.sha1
, tree_sha1
);
1320 it
->pcache
.tree_data
= data
;
1321 it
->pcache
.tree_size
= size
;
1324 static void cleanup_preferred_base(void)
1326 struct pbase_tree
*it
;
1332 struct pbase_tree
*this = it
;
1334 free(this->pcache
.tree_data
);
1338 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1339 if (!pbase_tree_cache
[i
])
1341 free(pbase_tree_cache
[i
]->tree_data
);
1342 free(pbase_tree_cache
[i
]);
1343 pbase_tree_cache
[i
] = NULL
;
1346 free(done_pbase_paths
);
1347 done_pbase_paths
= NULL
;
1348 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1351 static void check_object(struct object_entry
*entry
)
1353 if (entry
->in_pack
) {
1354 struct packed_git
*p
= entry
->in_pack
;
1355 struct pack_window
*w_curs
= NULL
;
1356 const unsigned char *base_ref
= NULL
;
1357 struct object_entry
*base_entry
;
1358 unsigned long used
, used_0
;
1359 unsigned long avail
;
1361 unsigned char *buf
, c
;
1363 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1366 * We want in_pack_type even if we do not reuse delta
1367 * since non-delta representations could still be reused.
1369 used
= unpack_object_header_buffer(buf
, avail
,
1370 &entry
->in_pack_type
,
1376 * Determine if this is a delta and if so whether we can
1377 * reuse it or not. Otherwise let's find out as cheaply as
1378 * possible what the actual type and size for this object is.
1380 switch (entry
->in_pack_type
) {
1382 /* Not a delta hence we've already got all we need. */
1383 entry
->type
= entry
->in_pack_type
;
1384 entry
->in_pack_header_size
= used
;
1385 if (entry
->type
< OBJ_COMMIT
|| entry
->type
> OBJ_BLOB
)
1387 unuse_pack(&w_curs
);
1390 if (reuse_delta
&& !entry
->preferred_base
)
1391 base_ref
= use_pack(p
, &w_curs
,
1392 entry
->in_pack_offset
+ used
, NULL
);
1393 entry
->in_pack_header_size
= used
+ 20;
1396 buf
= use_pack(p
, &w_curs
,
1397 entry
->in_pack_offset
+ used
, NULL
);
1403 if (!ofs
|| MSB(ofs
, 7)) {
1404 error("delta base offset overflow in pack for %s",
1405 sha1_to_hex(entry
->idx
.sha1
));
1409 ofs
= (ofs
<< 7) + (c
& 127);
1411 ofs
= entry
->in_pack_offset
- ofs
;
1412 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1413 error("delta base offset out of bound for %s",
1414 sha1_to_hex(entry
->idx
.sha1
));
1417 if (reuse_delta
&& !entry
->preferred_base
) {
1418 struct revindex_entry
*revidx
;
1419 revidx
= find_pack_revindex(p
, ofs
);
1422 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1424 entry
->in_pack_header_size
= used
+ used_0
;
1428 if (base_ref
&& (base_entry
= packlist_find(&to_pack
, base_ref
, NULL
))) {
1430 * If base_ref was set above that means we wish to
1431 * reuse delta data, and we even found that base
1432 * in the list of objects we want to pack. Goodie!
1434 * Depth value does not matter - find_deltas() will
1435 * never consider reused delta as the base object to
1436 * deltify other objects against, in order to avoid
1439 entry
->type
= entry
->in_pack_type
;
1440 entry
->delta
= base_entry
;
1441 entry
->delta_size
= entry
->size
;
1442 entry
->delta_sibling
= base_entry
->delta_child
;
1443 base_entry
->delta_child
= entry
;
1444 unuse_pack(&w_curs
);
1450 * This must be a delta and we already know what the
1451 * final object type is. Let's extract the actual
1452 * object size from the delta header.
1454 entry
->size
= get_size_from_delta(p
, &w_curs
,
1455 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1456 if (entry
->size
== 0)
1458 unuse_pack(&w_curs
);
1463 * No choice but to fall back to the recursive delta walk
1464 * with sha1_object_info() to find about the object type
1468 unuse_pack(&w_curs
);
1471 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1473 * The error condition is checked in prepare_pack(). This is
1474 * to permit a missing preferred base object to be ignored
1475 * as a preferred base. Doing so can result in a larger
1476 * pack file, but the transfer will still take place.
1480 static int pack_offset_sort(const void *_a
, const void *_b
)
1482 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1483 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1485 /* avoid filesystem trashing with loose objects */
1486 if (!a
->in_pack
&& !b
->in_pack
)
1487 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1489 if (a
->in_pack
< b
->in_pack
)
1491 if (a
->in_pack
> b
->in_pack
)
1493 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1494 (a
->in_pack_offset
> b
->in_pack_offset
);
1498 * Drop an on-disk delta we were planning to reuse. Naively, this would
1499 * just involve blanking out the "delta" field, but we have to deal
1500 * with some extra book-keeping:
1502 * 1. Removing ourselves from the delta_sibling linked list.
1504 * 2. Updating our size/type to the non-delta representation. These were
1505 * either not recorded initially (size) or overwritten with the delta type
1506 * (type) when check_object() decided to reuse the delta.
1508 static void drop_reused_delta(struct object_entry
*entry
)
1510 struct object_entry
**p
= &entry
->delta
->delta_child
;
1511 struct object_info oi
= OBJECT_INFO_INIT
;
1515 *p
= (*p
)->delta_sibling
;
1517 p
= &(*p
)->delta_sibling
;
1519 entry
->delta
= NULL
;
1521 oi
.sizep
= &entry
->size
;
1522 oi
.typep
= &entry
->type
;
1523 if (packed_object_info(entry
->in_pack
, entry
->in_pack_offset
, &oi
) < 0) {
1525 * We failed to get the info from this pack for some reason;
1526 * fall back to sha1_object_info, which may find another copy.
1527 * And if that fails, the error will be recorded in entry->type
1528 * and dealt with in prepare_pack().
1530 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1535 * Follow the chain of deltas from this entry onward, throwing away any links
1536 * that cause us to hit a cycle (as determined by the DFS state flags in
1539 static void break_delta_chains(struct object_entry
*entry
)
1541 /* If it's not a delta, it can't be part of a cycle. */
1542 if (!entry
->delta
) {
1543 entry
->dfs_state
= DFS_DONE
;
1547 switch (entry
->dfs_state
) {
1550 * This is the first time we've seen the object. We mark it as
1551 * part of the active potential cycle and recurse.
1553 entry
->dfs_state
= DFS_ACTIVE
;
1554 break_delta_chains(entry
->delta
);
1555 entry
->dfs_state
= DFS_DONE
;
1559 /* object already examined, and not part of a cycle */
1564 * We found a cycle that needs broken. It would be correct to
1565 * break any link in the chain, but it's convenient to
1568 drop_reused_delta(entry
);
1569 entry
->dfs_state
= DFS_DONE
;
1574 static void get_object_details(void)
1577 struct object_entry
**sorted_by_offset
;
1579 sorted_by_offset
= xcalloc(to_pack
.nr_objects
, sizeof(struct object_entry
*));
1580 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1581 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
1582 qsort(sorted_by_offset
, to_pack
.nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1584 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
1585 struct object_entry
*entry
= sorted_by_offset
[i
];
1586 check_object(entry
);
1587 if (big_file_threshold
< entry
->size
)
1588 entry
->no_try_delta
= 1;
1592 * This must happen in a second pass, since we rely on the delta
1593 * information for the whole list being completed.
1595 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1596 break_delta_chains(&to_pack
.objects
[i
]);
1598 free(sorted_by_offset
);
1602 * We search for deltas in a list sorted by type, by filename hash, and then
1603 * by size, so that we see progressively smaller and smaller files.
1604 * That's because we prefer deltas to be from the bigger file
1605 * to the smaller -- deletes are potentially cheaper, but perhaps
1606 * more importantly, the bigger file is likely the more recent
1607 * one. The deepest deltas are therefore the oldest objects which are
1608 * less susceptible to be accessed often.
1610 static int type_size_sort(const void *_a
, const void *_b
)
1612 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1613 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1615 if (a
->type
> b
->type
)
1617 if (a
->type
< b
->type
)
1619 if (a
->hash
> b
->hash
)
1621 if (a
->hash
< b
->hash
)
1623 if (a
->preferred_base
> b
->preferred_base
)
1625 if (a
->preferred_base
< b
->preferred_base
)
1627 if (a
->size
> b
->size
)
1629 if (a
->size
< b
->size
)
1631 return a
< b
? -1 : (a
> b
); /* newest first */
1635 struct object_entry
*entry
;
1637 struct delta_index
*index
;
1641 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1642 unsigned long delta_size
)
1644 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1647 if (delta_size
< cache_max_small_delta_size
)
1650 /* cache delta, if objects are large enough compared to delta size */
1651 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1659 static pthread_mutex_t read_mutex
;
1660 #define read_lock() pthread_mutex_lock(&read_mutex)
1661 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1663 static pthread_mutex_t cache_mutex
;
1664 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1665 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1667 static pthread_mutex_t progress_mutex
;
1668 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1669 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1673 #define read_lock() (void)0
1674 #define read_unlock() (void)0
1675 #define cache_lock() (void)0
1676 #define cache_unlock() (void)0
1677 #define progress_lock() (void)0
1678 #define progress_unlock() (void)0
1682 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1683 unsigned max_depth
, unsigned long *mem_usage
)
1685 struct object_entry
*trg_entry
= trg
->entry
;
1686 struct object_entry
*src_entry
= src
->entry
;
1687 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1689 enum object_type type
;
1692 /* Don't bother doing diffs between different types */
1693 if (trg_entry
->type
!= src_entry
->type
)
1697 * We do not bother to try a delta that we discarded on an
1698 * earlier try, but only when reusing delta data. Note that
1699 * src_entry that is marked as the preferred_base should always
1700 * be considered, as even if we produce a suboptimal delta against
1701 * it, we will still save the transfer cost, as we already know
1702 * the other side has it and we won't send src_entry at all.
1704 if (reuse_delta
&& trg_entry
->in_pack
&&
1705 trg_entry
->in_pack
== src_entry
->in_pack
&&
1706 !src_entry
->preferred_base
&&
1707 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1708 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1711 /* Let's not bust the allowed depth. */
1712 if (src
->depth
>= max_depth
)
1715 /* Now some size filtering heuristics. */
1716 trg_size
= trg_entry
->size
;
1717 if (!trg_entry
->delta
) {
1718 max_size
= trg_size
/2 - 20;
1721 max_size
= trg_entry
->delta_size
;
1722 ref_depth
= trg
->depth
;
1724 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
1725 (max_depth
- ref_depth
+ 1);
1728 src_size
= src_entry
->size
;
1729 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1730 if (sizediff
>= max_size
)
1732 if (trg_size
< src_size
/ 32)
1735 /* Load data if not already done */
1738 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1741 die("object %s cannot be read",
1742 sha1_to_hex(trg_entry
->idx
.sha1
));
1744 die("object %s inconsistent object length (%lu vs %lu)",
1745 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1750 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1753 if (src_entry
->preferred_base
) {
1754 static int warned
= 0;
1756 warning("object %s cannot be read",
1757 sha1_to_hex(src_entry
->idx
.sha1
));
1759 * Those objects are not included in the
1760 * resulting pack. Be resilient and ignore
1761 * them if they can't be read, in case the
1762 * pack could be created nevertheless.
1766 die("object %s cannot be read",
1767 sha1_to_hex(src_entry
->idx
.sha1
));
1770 die("object %s inconsistent object length (%lu vs %lu)",
1771 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1775 src
->index
= create_delta_index(src
->data
, src_size
);
1777 static int warned
= 0;
1779 warning("suboptimal pack - out of memory");
1782 *mem_usage
+= sizeof_delta_index(src
->index
);
1785 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1789 if (trg_entry
->delta
) {
1790 /* Prefer only shallower same-sized deltas. */
1791 if (delta_size
== trg_entry
->delta_size
&&
1792 src
->depth
+ 1 >= trg
->depth
) {
1799 * Handle memory allocation outside of the cache
1800 * accounting lock. Compiler will optimize the strangeness
1801 * away when NO_PTHREADS is defined.
1803 free(trg_entry
->delta_data
);
1805 if (trg_entry
->delta_data
) {
1806 delta_cache_size
-= trg_entry
->delta_size
;
1807 trg_entry
->delta_data
= NULL
;
1809 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1810 delta_cache_size
+= delta_size
;
1812 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1818 trg_entry
->delta
= src_entry
;
1819 trg_entry
->delta_size
= delta_size
;
1820 trg
->depth
= src
->depth
+ 1;
1825 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1827 struct object_entry
*child
= me
->delta_child
;
1830 unsigned int c
= check_delta_limit(child
, n
+ 1);
1833 child
= child
->delta_sibling
;
1838 static unsigned long free_unpacked(struct unpacked
*n
)
1840 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1841 free_delta_index(n
->index
);
1844 freed_mem
+= n
->entry
->size
;
1853 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1854 int window
, int depth
, unsigned *processed
)
1856 uint32_t i
, idx
= 0, count
= 0;
1857 struct unpacked
*array
;
1858 unsigned long mem_usage
= 0;
1860 array
= xcalloc(window
, sizeof(struct unpacked
));
1863 struct object_entry
*entry
;
1864 struct unpacked
*n
= array
+ idx
;
1865 int j
, max_depth
, best_base
= -1;
1874 if (!entry
->preferred_base
) {
1876 display_progress(progress_state
, *processed
);
1880 mem_usage
-= free_unpacked(n
);
1883 while (window_memory_limit
&&
1884 mem_usage
> window_memory_limit
&&
1886 uint32_t tail
= (idx
+ window
- count
) % window
;
1887 mem_usage
-= free_unpacked(array
+ tail
);
1891 /* We do not compute delta to *create* objects we are not
1894 if (entry
->preferred_base
)
1898 * If the current object is at pack edge, take the depth the
1899 * objects that depend on the current object into account
1900 * otherwise they would become too deep.
1903 if (entry
->delta_child
) {
1904 max_depth
-= check_delta_limit(entry
, 0);
1912 uint32_t other_idx
= idx
+ j
;
1914 if (other_idx
>= window
)
1915 other_idx
-= window
;
1916 m
= array
+ other_idx
;
1919 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1923 best_base
= other_idx
;
1927 * If we decided to cache the delta data, then it is best
1928 * to compress it right away. First because we have to do
1929 * it anyway, and doing it here while we're threaded will
1930 * save a lot of time in the non threaded write phase,
1931 * as well as allow for caching more deltas within
1932 * the same cache size limit.
1934 * But only if not writing to stdout, since in that case
1935 * the network is most likely throttling writes anyway,
1936 * and therefore it is best to go to the write phase ASAP
1937 * instead, as we can afford spending more time compressing
1938 * between writes at that moment.
1940 if (entry
->delta_data
&& !pack_to_stdout
) {
1941 entry
->z_delta_size
= do_compress(&entry
->delta_data
,
1944 delta_cache_size
-= entry
->delta_size
;
1945 delta_cache_size
+= entry
->z_delta_size
;
1949 /* if we made n a delta, and if n is already at max
1950 * depth, leaving it in the window is pointless. we
1951 * should evict it first.
1953 if (entry
->delta
&& max_depth
<= n
->depth
)
1957 * Move the best delta base up in the window, after the
1958 * currently deltified object, to keep it longer. It will
1959 * be the first base object to be attempted next.
1962 struct unpacked swap
= array
[best_base
];
1963 int dist
= (window
+ idx
- best_base
) % window
;
1964 int dst
= best_base
;
1966 int src
= (dst
+ 1) % window
;
1967 array
[dst
] = array
[src
];
1975 if (count
+ 1 < window
)
1981 for (i
= 0; i
< window
; ++i
) {
1982 free_delta_index(array
[i
].index
);
1983 free(array
[i
].data
);
1990 static void try_to_free_from_threads(size_t size
)
1993 release_pack_memory(size
);
1997 static try_to_free_t old_try_to_free_routine
;
2000 * The main thread waits on the condition that (at least) one of the workers
2001 * has stopped working (which is indicated in the .working member of
2002 * struct thread_params).
2003 * When a work thread has completed its work, it sets .working to 0 and
2004 * signals the main thread and waits on the condition that .data_ready
2008 struct thread_params
{
2010 struct object_entry
**list
;
2017 pthread_mutex_t mutex
;
2018 pthread_cond_t cond
;
2019 unsigned *processed
;
2022 static pthread_cond_t progress_cond
;
2025 * Mutex and conditional variable can't be statically-initialized on Windows.
2027 static void init_threaded_search(void)
2029 init_recursive_mutex(&read_mutex
);
2030 pthread_mutex_init(&cache_mutex
, NULL
);
2031 pthread_mutex_init(&progress_mutex
, NULL
);
2032 pthread_cond_init(&progress_cond
, NULL
);
2033 old_try_to_free_routine
= set_try_to_free_routine(try_to_free_from_threads
);
2036 static void cleanup_threaded_search(void)
2038 set_try_to_free_routine(old_try_to_free_routine
);
2039 pthread_cond_destroy(&progress_cond
);
2040 pthread_mutex_destroy(&read_mutex
);
2041 pthread_mutex_destroy(&cache_mutex
);
2042 pthread_mutex_destroy(&progress_mutex
);
2045 static void *threaded_find_deltas(void *arg
)
2047 struct thread_params
*me
= arg
;
2049 while (me
->remaining
) {
2050 find_deltas(me
->list
, &me
->remaining
,
2051 me
->window
, me
->depth
, me
->processed
);
2055 pthread_cond_signal(&progress_cond
);
2059 * We must not set ->data_ready before we wait on the
2060 * condition because the main thread may have set it to 1
2061 * before we get here. In order to be sure that new
2062 * work is available if we see 1 in ->data_ready, it
2063 * was initialized to 0 before this thread was spawned
2064 * and we reset it to 0 right away.
2066 pthread_mutex_lock(&me
->mutex
);
2067 while (!me
->data_ready
)
2068 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2070 pthread_mutex_unlock(&me
->mutex
);
2072 /* leave ->working 1 so that this doesn't get more work assigned */
2076 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2077 int window
, int depth
, unsigned *processed
)
2079 struct thread_params
*p
;
2080 int i
, ret
, active_threads
= 0;
2082 init_threaded_search();
2084 if (delta_search_threads
<= 1) {
2085 find_deltas(list
, &list_size
, window
, depth
, processed
);
2086 cleanup_threaded_search();
2089 if (progress
> pack_to_stdout
)
2090 fprintf(stderr
, "Delta compression using up to %d threads.\n",
2091 delta_search_threads
);
2092 p
= xcalloc(delta_search_threads
, sizeof(*p
));
2094 /* Partition the work amongst work threads. */
2095 for (i
= 0; i
< delta_search_threads
; i
++) {
2096 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2098 /* don't use too small segments or no deltas will be found */
2099 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2102 p
[i
].window
= window
;
2104 p
[i
].processed
= processed
;
2106 p
[i
].data_ready
= 0;
2108 /* try to split chunks on "path" boundaries */
2109 while (sub_size
&& sub_size
< list_size
&&
2110 list
[sub_size
]->hash
&&
2111 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2115 p
[i
].list_size
= sub_size
;
2116 p
[i
].remaining
= sub_size
;
2119 list_size
-= sub_size
;
2122 /* Start work threads. */
2123 for (i
= 0; i
< delta_search_threads
; i
++) {
2124 if (!p
[i
].list_size
)
2126 pthread_mutex_init(&p
[i
].mutex
, NULL
);
2127 pthread_cond_init(&p
[i
].cond
, NULL
);
2128 ret
= pthread_create(&p
[i
].thread
, NULL
,
2129 threaded_find_deltas
, &p
[i
]);
2131 die("unable to create thread: %s", strerror(ret
));
2136 * Now let's wait for work completion. Each time a thread is done
2137 * with its work, we steal half of the remaining work from the
2138 * thread with the largest number of unprocessed objects and give
2139 * it to that newly idle thread. This ensure good load balancing
2140 * until the remaining object list segments are simply too short
2141 * to be worth splitting anymore.
2143 while (active_threads
) {
2144 struct thread_params
*target
= NULL
;
2145 struct thread_params
*victim
= NULL
;
2146 unsigned sub_size
= 0;
2150 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
2155 pthread_cond_wait(&progress_cond
, &progress_mutex
);
2158 for (i
= 0; i
< delta_search_threads
; i
++)
2159 if (p
[i
].remaining
> 2*window
&&
2160 (!victim
|| victim
->remaining
< p
[i
].remaining
))
2163 sub_size
= victim
->remaining
/ 2;
2164 list
= victim
->list
+ victim
->list_size
- sub_size
;
2165 while (sub_size
&& list
[0]->hash
&&
2166 list
[0]->hash
== list
[-1]->hash
) {
2172 * It is possible for some "paths" to have
2173 * so many objects that no hash boundary
2174 * might be found. Let's just steal the
2175 * exact half in that case.
2177 sub_size
= victim
->remaining
/ 2;
2180 target
->list
= list
;
2181 victim
->list_size
-= sub_size
;
2182 victim
->remaining
-= sub_size
;
2184 target
->list_size
= sub_size
;
2185 target
->remaining
= sub_size
;
2186 target
->working
= 1;
2189 pthread_mutex_lock(&target
->mutex
);
2190 target
->data_ready
= 1;
2191 pthread_cond_signal(&target
->cond
);
2192 pthread_mutex_unlock(&target
->mutex
);
2195 pthread_join(target
->thread
, NULL
);
2196 pthread_cond_destroy(&target
->cond
);
2197 pthread_mutex_destroy(&target
->mutex
);
2201 cleanup_threaded_search();
2206 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2209 static int add_ref_tag(const char *path
, const struct object_id
*oid
, int flag
, void *cb_data
)
2211 struct object_id peeled
;
2213 if (starts_with(path
, "refs/tags/") && /* is a tag? */
2214 !peel_ref(path
, peeled
.hash
) && /* peelable? */
2215 packlist_find(&to_pack
, peeled
.hash
, NULL
)) /* object packed? */
2216 add_object_entry(oid
->hash
, OBJ_TAG
, NULL
, 0);
2220 static void prepare_pack(int window
, int depth
)
2222 struct object_entry
**delta_list
;
2223 uint32_t i
, nr_deltas
;
2226 get_object_details();
2229 * If we're locally repacking then we need to be doubly careful
2230 * from now on in order to make sure no stealth corruption gets
2231 * propagated to the new pack. Clients receiving streamed packs
2232 * should validate everything they get anyway so no need to incur
2233 * the additional cost here in that case.
2235 if (!pack_to_stdout
)
2236 do_check_packed_object_crc
= 1;
2238 if (!to_pack
.nr_objects
|| !window
|| !depth
)
2241 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
2244 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2245 struct object_entry
*entry
= to_pack
.objects
+ i
;
2248 /* This happens if we decided to reuse existing
2249 * delta from a pack. "reuse_delta &&" is implied.
2253 if (entry
->size
< 50)
2256 if (entry
->no_try_delta
)
2259 if (!entry
->preferred_base
) {
2261 if (entry
->type
< 0)
2262 die("unable to get type of object %s",
2263 sha1_to_hex(entry
->idx
.sha1
));
2265 if (entry
->type
< 0) {
2267 * This object is not found, but we
2268 * don't have to include it anyway.
2274 delta_list
[n
++] = entry
;
2277 if (nr_deltas
&& n
> 1) {
2278 unsigned nr_done
= 0;
2280 progress_state
= start_progress(_("Compressing objects"),
2282 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
2283 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
2284 stop_progress(&progress_state
);
2285 if (nr_done
!= nr_deltas
)
2286 die("inconsistency with delta count");
2291 static int git_pack_config(const char *k
, const char *v
, void *cb
)
2293 if (!strcmp(k
, "pack.window")) {
2294 window
= git_config_int(k
, v
);
2297 if (!strcmp(k
, "pack.windowmemory")) {
2298 window_memory_limit
= git_config_ulong(k
, v
);
2301 if (!strcmp(k
, "pack.depth")) {
2302 depth
= git_config_int(k
, v
);
2305 if (!strcmp(k
, "pack.compression")) {
2306 int level
= git_config_int(k
, v
);
2308 level
= Z_DEFAULT_COMPRESSION
;
2309 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
2310 die("bad pack compression level %d", level
);
2311 pack_compression_level
= level
;
2312 pack_compression_seen
= 1;
2315 if (!strcmp(k
, "pack.deltacachesize")) {
2316 max_delta_cache_size
= git_config_int(k
, v
);
2319 if (!strcmp(k
, "pack.deltacachelimit")) {
2320 cache_max_small_delta_size
= git_config_int(k
, v
);
2323 if (!strcmp(k
, "pack.writebitmaphashcache")) {
2324 if (git_config_bool(k
, v
))
2325 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
2327 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
2329 if (!strcmp(k
, "pack.usebitmaps")) {
2330 use_bitmap_index
= git_config_bool(k
, v
);
2333 if (!strcmp(k
, "pack.threads")) {
2334 delta_search_threads
= git_config_int(k
, v
);
2335 if (delta_search_threads
< 0)
2336 die("invalid number of threads specified (%d)",
2337 delta_search_threads
);
2339 if (delta_search_threads
!= 1)
2340 warning("no threads support, ignoring %s", k
);
2344 if (!strcmp(k
, "pack.indexversion")) {
2345 pack_idx_opts
.version
= git_config_int(k
, v
);
2346 if (pack_idx_opts
.version
> 2)
2347 die("bad pack.indexversion=%"PRIu32
,
2348 pack_idx_opts
.version
);
2351 return git_default_config(k
, v
, cb
);
2354 static void read_object_list_from_stdin(void)
2356 char line
[40 + 1 + PATH_MAX
+ 2];
2357 unsigned char sha1
[20];
2360 if (!fgets(line
, sizeof(line
), stdin
)) {
2364 die("fgets returned NULL, not EOF, not error!");
2370 if (line
[0] == '-') {
2371 if (get_sha1_hex(line
+1, sha1
))
2372 die("expected edge sha1, got garbage:\n %s",
2374 add_preferred_base(sha1
);
2377 if (get_sha1_hex(line
, sha1
))
2378 die("expected sha1, got garbage:\n %s", line
);
2380 add_preferred_base_object(line
+41);
2381 add_object_entry(sha1
, 0, line
+41, 0);
2385 #define OBJECT_ADDED (1u<<20)
2387 static void show_commit(struct commit
*commit
, void *data
)
2389 add_object_entry(commit
->object
.oid
.hash
, OBJ_COMMIT
, NULL
, 0);
2390 commit
->object
.flags
|= OBJECT_ADDED
;
2392 if (write_bitmap_index
)
2393 index_commit_for_bitmap(commit
);
2396 static void show_object(struct object
*obj
, const char *name
, void *data
)
2398 add_preferred_base_object(name
);
2399 add_object_entry(obj
->oid
.hash
, obj
->type
, name
, 0);
2400 obj
->flags
|= OBJECT_ADDED
;
2403 static void show_edge(struct commit
*commit
)
2405 add_preferred_base(commit
->object
.oid
.hash
);
2408 struct in_pack_object
{
2410 struct object
*object
;
2416 struct in_pack_object
*array
;
2419 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
2421 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->oid
.hash
, p
);
2422 in_pack
->array
[in_pack
->nr
].object
= object
;
2427 * Compare the objects in the offset order, in order to emulate the
2428 * "git rev-list --objects" output that produced the pack originally.
2430 static int ofscmp(const void *a_
, const void *b_
)
2432 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
2433 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
2435 if (a
->offset
< b
->offset
)
2437 else if (a
->offset
> b
->offset
)
2440 return oidcmp(&a
->object
->oid
, &b
->object
->oid
);
2443 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
2445 struct packed_git
*p
;
2446 struct in_pack in_pack
;
2449 memset(&in_pack
, 0, sizeof(in_pack
));
2451 for (p
= packed_git
; p
; p
= p
->next
) {
2452 const unsigned char *sha1
;
2455 if (!p
->pack_local
|| p
->pack_keep
)
2457 if (open_pack_index(p
))
2458 die("cannot open pack index");
2460 ALLOC_GROW(in_pack
.array
,
2461 in_pack
.nr
+ p
->num_objects
,
2464 for (i
= 0; i
< p
->num_objects
; i
++) {
2465 sha1
= nth_packed_object_sha1(p
, i
);
2466 o
= lookup_unknown_object(sha1
);
2467 if (!(o
->flags
& OBJECT_ADDED
))
2468 mark_in_pack_object(o
, p
, &in_pack
);
2469 o
->flags
|= OBJECT_ADDED
;
2474 qsort(in_pack
.array
, in_pack
.nr
, sizeof(in_pack
.array
[0]),
2476 for (i
= 0; i
< in_pack
.nr
; i
++) {
2477 struct object
*o
= in_pack
.array
[i
].object
;
2478 add_object_entry(o
->oid
.hash
, o
->type
, "", 0);
2481 free(in_pack
.array
);
2484 static int add_loose_object(const unsigned char *sha1
, const char *path
,
2487 enum object_type type
= sha1_object_info(sha1
, NULL
);
2490 warning("loose object at %s could not be examined", path
);
2494 add_object_entry(sha1
, type
, "", 0);
2499 * We actually don't even have to worry about reachability here.
2500 * add_object_entry will weed out duplicates, so we just add every
2501 * loose object we find.
2503 static void add_unreachable_loose_objects(void)
2505 for_each_loose_file_in_objdir(get_object_directory(),
2510 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1
)
2512 static struct packed_git
*last_found
= (void *)1;
2513 struct packed_git
*p
;
2515 p
= (last_found
!= (void *)1) ? last_found
: packed_git
;
2518 if ((!p
->pack_local
|| p
->pack_keep
) &&
2519 find_pack_entry_one(sha1
, p
)) {
2523 if (p
== last_found
)
2527 if (p
== last_found
)
2534 * Store a list of sha1s that are should not be discarded
2535 * because they are either written too recently, or are
2536 * reachable from another object that was.
2538 * This is filled by get_object_list.
2540 static struct sha1_array recent_objects
;
2542 static int loosened_object_can_be_discarded(const unsigned char *sha1
,
2543 unsigned long mtime
)
2545 if (!unpack_unreachable_expiration
)
2547 if (mtime
> unpack_unreachable_expiration
)
2549 if (sha1_array_lookup(&recent_objects
, sha1
) >= 0)
2554 static void loosen_unused_packed_objects(struct rev_info
*revs
)
2556 struct packed_git
*p
;
2558 const unsigned char *sha1
;
2560 for (p
= packed_git
; p
; p
= p
->next
) {
2561 if (!p
->pack_local
|| p
->pack_keep
)
2564 if (open_pack_index(p
))
2565 die("cannot open pack index");
2567 for (i
= 0; i
< p
->num_objects
; i
++) {
2568 sha1
= nth_packed_object_sha1(p
, i
);
2569 if (!packlist_find(&to_pack
, sha1
, NULL
) &&
2570 !has_sha1_pack_kept_or_nonlocal(sha1
) &&
2571 !loosened_object_can_be_discarded(sha1
, p
->mtime
))
2572 if (force_object_loose(sha1
, p
->mtime
))
2573 die("unable to force loose object");
2579 * This tracks any options which a reader of the pack might
2580 * not understand, and which would therefore prevent blind reuse
2581 * of what we have on disk.
2583 static int pack_options_allow_reuse(void)
2585 return allow_ofs_delta
;
2588 static int get_object_list_from_bitmap(struct rev_info
*revs
)
2590 if (prepare_bitmap_walk(revs
) < 0)
2593 if (pack_options_allow_reuse() &&
2594 !reuse_partial_packfile_from_bitmap(
2596 &reuse_packfile_objects
,
2597 &reuse_packfile_offset
)) {
2598 assert(reuse_packfile_objects
);
2599 nr_result
+= reuse_packfile_objects
;
2600 display_progress(progress_state
, nr_result
);
2603 traverse_bitmap_commit_list(&add_object_entry_from_bitmap
);
2607 static void record_recent_object(struct object
*obj
,
2611 sha1_array_append(&recent_objects
, obj
->oid
.hash
);
2614 static void record_recent_commit(struct commit
*commit
, void *data
)
2616 sha1_array_append(&recent_objects
, commit
->object
.oid
.hash
);
2619 static void get_object_list(int ac
, const char **av
)
2621 struct rev_info revs
;
2625 init_revisions(&revs
, NULL
);
2626 save_commit_buffer
= 0;
2627 setup_revisions(ac
, av
, &revs
, NULL
);
2629 /* make sure shallows are read */
2630 is_repository_shallow();
2632 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2633 int len
= strlen(line
);
2634 if (len
&& line
[len
- 1] == '\n')
2639 if (!strcmp(line
, "--not")) {
2640 flags
^= UNINTERESTING
;
2641 write_bitmap_index
= 0;
2644 if (starts_with(line
, "--shallow ")) {
2645 unsigned char sha1
[20];
2646 if (get_sha1_hex(line
+ 10, sha1
))
2647 die("not an SHA-1 '%s'", line
+ 10);
2648 register_shallow(sha1
);
2649 use_bitmap_index
= 0;
2652 die("not a rev '%s'", line
);
2654 if (handle_revision_arg(line
, &revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
2655 die("bad revision '%s'", line
);
2658 if (use_bitmap_index
&& !get_object_list_from_bitmap(&revs
))
2661 if (prepare_revision_walk(&revs
))
2662 die("revision walk setup failed");
2663 mark_edges_uninteresting(&revs
, show_edge
);
2664 traverse_commit_list(&revs
, show_commit
, show_object
, NULL
);
2666 if (unpack_unreachable_expiration
) {
2667 revs
.ignore_missing_links
= 1;
2668 if (add_unseen_recent_objects_to_traversal(&revs
,
2669 unpack_unreachable_expiration
))
2670 die("unable to add recent objects");
2671 if (prepare_revision_walk(&revs
))
2672 die("revision walk setup failed");
2673 traverse_commit_list(&revs
, record_recent_commit
,
2674 record_recent_object
, NULL
);
2677 if (keep_unreachable
)
2678 add_objects_in_unpacked_packs(&revs
);
2679 if (pack_loose_unreachable
)
2680 add_unreachable_loose_objects();
2681 if (unpack_unreachable
)
2682 loosen_unused_packed_objects(&revs
);
2684 sha1_array_clear(&recent_objects
);
2687 static int option_parse_index_version(const struct option
*opt
,
2688 const char *arg
, int unset
)
2691 const char *val
= arg
;
2692 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
2693 if (pack_idx_opts
.version
> 2)
2694 die(_("unsupported index version %s"), val
);
2695 if (*c
== ',' && c
[1])
2696 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
2697 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
2698 die(_("bad index version '%s'"), val
);
2702 static int option_parse_unpack_unreachable(const struct option
*opt
,
2703 const char *arg
, int unset
)
2706 unpack_unreachable
= 0;
2707 unpack_unreachable_expiration
= 0;
2710 unpack_unreachable
= 1;
2712 unpack_unreachable_expiration
= approxidate(arg
);
2717 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
2719 int use_internal_rev_list
= 0;
2722 int all_progress_implied
= 0;
2723 struct argv_array rp
= ARGV_ARRAY_INIT
;
2724 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
2725 int rev_list_index
= 0;
2726 struct option pack_objects_options
[] = {
2727 OPT_SET_INT('q', "quiet", &progress
,
2728 N_("do not show progress meter"), 0),
2729 OPT_SET_INT(0, "progress", &progress
,
2730 N_("show progress meter"), 1),
2731 OPT_SET_INT(0, "all-progress", &progress
,
2732 N_("show progress meter during object writing phase"), 2),
2733 OPT_BOOL(0, "all-progress-implied",
2734 &all_progress_implied
,
2735 N_("similar to --all-progress when progress meter is shown")),
2736 { OPTION_CALLBACK
, 0, "index-version", NULL
, N_("version[,offset]"),
2737 N_("write the pack index file in the specified idx format version"),
2738 0, option_parse_index_version
},
2739 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
2740 N_("maximum size of each output pack file")),
2741 OPT_BOOL(0, "local", &local
,
2742 N_("ignore borrowed objects from alternate object store")),
2743 OPT_BOOL(0, "incremental", &incremental
,
2744 N_("ignore packed objects")),
2745 OPT_INTEGER(0, "window", &window
,
2746 N_("limit pack window by objects")),
2747 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
2748 N_("limit pack window by memory in addition to object limit")),
2749 OPT_INTEGER(0, "depth", &depth
,
2750 N_("maximum length of delta chain allowed in the resulting pack")),
2751 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
2752 N_("reuse existing deltas")),
2753 OPT_BOOL(0, "reuse-object", &reuse_object
,
2754 N_("reuse existing objects")),
2755 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
2756 N_("use OFS_DELTA objects")),
2757 OPT_INTEGER(0, "threads", &delta_search_threads
,
2758 N_("use threads when searching for best delta matches")),
2759 OPT_BOOL(0, "non-empty", &non_empty
,
2760 N_("do not create an empty pack output")),
2761 OPT_BOOL(0, "revs", &use_internal_rev_list
,
2762 N_("read revision arguments from standard input")),
2763 { OPTION_SET_INT
, 0, "unpacked", &rev_list_unpacked
, NULL
,
2764 N_("limit the objects to those that are not yet packed"),
2765 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2766 { OPTION_SET_INT
, 0, "all", &rev_list_all
, NULL
,
2767 N_("include objects reachable from any reference"),
2768 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2769 { OPTION_SET_INT
, 0, "reflog", &rev_list_reflog
, NULL
,
2770 N_("include objects referred by reflog entries"),
2771 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2772 { OPTION_SET_INT
, 0, "indexed-objects", &rev_list_index
, NULL
,
2773 N_("include objects referred to by the index"),
2774 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2775 OPT_BOOL(0, "stdout", &pack_to_stdout
,
2776 N_("output pack to stdout")),
2777 OPT_BOOL(0, "include-tag", &include_tag
,
2778 N_("include tag objects that refer to objects to be packed")),
2779 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
2780 N_("keep unreachable objects")),
2781 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
2782 N_("pack loose unreachable objects")),
2783 { OPTION_CALLBACK
, 0, "unpack-unreachable", NULL
, N_("time"),
2784 N_("unpack unreachable objects newer than <time>"),
2785 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
},
2786 OPT_BOOL(0, "thin", &thin
,
2787 N_("create thin packs")),
2788 OPT_BOOL(0, "shallow", &shallow
,
2789 N_("create packs suitable for shallow fetches")),
2790 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep
,
2791 N_("ignore packs that have companion .keep file")),
2792 OPT_INTEGER(0, "compression", &pack_compression_level
,
2793 N_("pack compression level")),
2794 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
2795 N_("do not hide commits by grafts"), 0),
2796 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
2797 N_("use a bitmap index if available to speed up counting objects")),
2798 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index
,
2799 N_("write a bitmap index together with the pack index")),
2803 check_replace_refs
= 0;
2805 reset_pack_idx_option(&pack_idx_opts
);
2806 git_config(git_pack_config
, NULL
);
2807 if (!pack_compression_seen
&& core_compression_seen
)
2808 pack_compression_level
= core_compression_level
;
2810 progress
= isatty(2);
2811 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
2815 base_name
= argv
[0];
2818 if (pack_to_stdout
!= !base_name
|| argc
)
2819 usage_with_options(pack_usage
, pack_objects_options
);
2821 argv_array_push(&rp
, "pack-objects");
2823 use_internal_rev_list
= 1;
2824 argv_array_push(&rp
, shallow
2825 ? "--objects-edge-aggressive"
2826 : "--objects-edge");
2828 argv_array_push(&rp
, "--objects");
2831 use_internal_rev_list
= 1;
2832 argv_array_push(&rp
, "--all");
2834 if (rev_list_reflog
) {
2835 use_internal_rev_list
= 1;
2836 argv_array_push(&rp
, "--reflog");
2838 if (rev_list_index
) {
2839 use_internal_rev_list
= 1;
2840 argv_array_push(&rp
, "--indexed-objects");
2842 if (rev_list_unpacked
) {
2843 use_internal_rev_list
= 1;
2844 argv_array_push(&rp
, "--unpacked");
2849 if (pack_compression_level
== -1)
2850 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
2851 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
2852 die("bad pack compression level %d", pack_compression_level
);
2854 if (!delta_search_threads
) /* --threads=0 means autodetect */
2855 delta_search_threads
= online_cpus();
2858 if (delta_search_threads
!= 1)
2859 warning("no threads support, ignoring --threads");
2861 if (!pack_to_stdout
&& !pack_size_limit
)
2862 pack_size_limit
= pack_size_limit_cfg
;
2863 if (pack_to_stdout
&& pack_size_limit
)
2864 die("--max-pack-size cannot be used to build a pack for transfer.");
2865 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
2866 warning("minimum pack size limit is 1 MiB");
2867 pack_size_limit
= 1024*1024;
2870 if (!pack_to_stdout
&& thin
)
2871 die("--thin cannot be used to build an indexable pack.");
2873 if (keep_unreachable
&& unpack_unreachable
)
2874 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2875 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
2876 unpack_unreachable_expiration
= 0;
2878 if (!use_internal_rev_list
|| !pack_to_stdout
|| is_repository_shallow())
2879 use_bitmap_index
= 0;
2881 if (pack_to_stdout
|| !rev_list_all
)
2882 write_bitmap_index
= 0;
2884 if (progress
&& all_progress_implied
)
2887 prepare_packed_git();
2888 if (ignore_packed_keep
) {
2889 struct packed_git
*p
;
2890 for (p
= packed_git
; p
; p
= p
->next
)
2891 if (p
->pack_local
&& p
->pack_keep
)
2893 if (!p
) /* no keep-able packs found */
2894 ignore_packed_keep
= 0;
2898 * unlike ignore_packed_keep above, we do not want to
2899 * unset "local" based on looking at packs, as it
2900 * also covers non-local objects
2902 struct packed_git
*p
;
2903 for (p
= packed_git
; p
; p
= p
->next
) {
2904 if (!p
->pack_local
) {
2905 have_non_local_packs
= 1;
2912 progress_state
= start_progress(_("Counting objects"), 0);
2913 if (!use_internal_rev_list
)
2914 read_object_list_from_stdin();
2916 get_object_list(rp
.argc
, rp
.argv
);
2917 argv_array_clear(&rp
);
2919 cleanup_preferred_base();
2920 if (include_tag
&& nr_result
)
2921 for_each_ref(add_ref_tag
, NULL
);
2922 stop_progress(&progress_state
);
2924 if (non_empty
&& !nr_result
)
2927 prepare_pack(window
, depth
);
2930 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
2931 " reused %"PRIu32
" (delta %"PRIu32
")\n",
2932 written
, written_delta
, reused
, reused_delta
);