3 #include "repository.h"
13 #include "pack-revindex.h"
14 #include "csum-file.h"
15 #include "tree-walk.h"
18 #include "list-objects.h"
19 #include "list-objects-filter.h"
20 #include "list-objects-filter-options.h"
21 #include "pack-objects.h"
24 #include "streaming.h"
25 #include "thread-utils.h"
26 #include "pack-bitmap.h"
27 #include "reachable.h"
28 #include "sha1-array.h"
29 #include "argv-array.h"
32 #include "object-store.h"
35 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
36 #define SIZE(obj) oe_size(&to_pack, obj)
37 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
38 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
39 #define DELTA(obj) oe_delta(&to_pack, obj)
40 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
41 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
42 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
43 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
44 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
45 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
47 static const char *pack_usage
[] = {
48 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
49 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
54 * Objects we are going to pack are collected in the `to_pack` structure.
55 * It contains an array (dynamically expanded) of the object data, and a map
56 * that can resolve SHA1s to their position in the array.
58 static struct packing_data to_pack
;
60 static struct pack_idx_entry
**written_list
;
61 static uint32_t nr_result
, nr_written
, nr_seen
;
64 static int reuse_delta
= 1, reuse_object
= 1;
65 static int keep_unreachable
, unpack_unreachable
, include_tag
;
66 static timestamp_t unpack_unreachable_expiration
;
67 static int pack_loose_unreachable
;
69 static int have_non_local_packs
;
70 static int incremental
;
71 static int ignore_packed_keep_on_disk
;
72 static int ignore_packed_keep_in_core
;
73 static int allow_ofs_delta
;
74 static struct pack_idx_option pack_idx_opts
;
75 static const char *base_name
;
76 static int progress
= 1;
77 static int window
= 10;
78 static unsigned long pack_size_limit
;
79 static int depth
= 50;
80 static int delta_search_threads
;
81 static int pack_to_stdout
;
82 static int num_preferred_base
;
83 static struct progress
*progress_state
;
85 static struct packed_git
*reuse_packfile
;
86 static uint32_t reuse_packfile_objects
;
87 static off_t reuse_packfile_offset
;
89 static int use_bitmap_index_default
= 1;
90 static int use_bitmap_index
= -1;
91 static int write_bitmap_index
;
92 static uint16_t write_bitmap_options
;
94 static int exclude_promisor_objects
;
96 static unsigned long delta_cache_size
= 0;
97 static unsigned long max_delta_cache_size
= DEFAULT_DELTA_CACHE_SIZE
;
98 static unsigned long cache_max_small_delta_size
= 1000;
100 static unsigned long window_memory_limit
= 0;
102 static struct list_objects_filter_options filter_options
;
104 enum missing_action
{
105 MA_ERROR
= 0, /* fail if any missing objects are encountered */
106 MA_ALLOW_ANY
, /* silently allow ALL missing objects */
107 MA_ALLOW_PROMISOR
, /* silently allow all missing PROMISOR objects */
109 static enum missing_action arg_missing_action
;
110 static show_object_fn fn_show_object
;
115 static uint32_t written
, written_delta
;
116 static uint32_t reused
, reused_delta
;
121 static struct commit
**indexed_commits
;
122 static unsigned int indexed_commits_nr
;
123 static unsigned int indexed_commits_alloc
;
125 static void index_commit_for_bitmap(struct commit
*commit
)
127 if (indexed_commits_nr
>= indexed_commits_alloc
) {
128 indexed_commits_alloc
= (indexed_commits_alloc
+ 32) * 2;
129 REALLOC_ARRAY(indexed_commits
, indexed_commits_alloc
);
132 indexed_commits
[indexed_commits_nr
++] = commit
;
135 static void *get_delta(struct object_entry
*entry
)
137 unsigned long size
, base_size
, delta_size
;
138 void *buf
, *base_buf
, *delta_buf
;
139 enum object_type type
;
141 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
143 die("unable to read %s", oid_to_hex(&entry
->idx
.oid
));
144 base_buf
= read_object_file(&DELTA(entry
)->idx
.oid
, &type
,
147 die("unable to read %s",
148 oid_to_hex(&DELTA(entry
)->idx
.oid
));
149 delta_buf
= diff_delta(base_buf
, base_size
,
150 buf
, size
, &delta_size
, 0);
151 if (!delta_buf
|| delta_size
!= DELTA_SIZE(entry
))
152 die("delta size changed");
158 static unsigned long do_compress(void **pptr
, unsigned long size
)
162 unsigned long maxsize
;
164 git_deflate_init(&stream
, pack_compression_level
);
165 maxsize
= git_deflate_bound(&stream
, size
);
168 out
= xmalloc(maxsize
);
172 stream
.avail_in
= size
;
173 stream
.next_out
= out
;
174 stream
.avail_out
= maxsize
;
175 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
177 git_deflate_end(&stream
);
180 return stream
.total_out
;
183 static unsigned long write_large_blob_data(struct git_istream
*st
, struct hashfile
*f
,
184 const struct object_id
*oid
)
187 unsigned char ibuf
[1024 * 16];
188 unsigned char obuf
[1024 * 16];
189 unsigned long olen
= 0;
191 git_deflate_init(&stream
, pack_compression_level
);
196 readlen
= read_istream(st
, ibuf
, sizeof(ibuf
));
198 die(_("unable to read %s"), oid_to_hex(oid
));
200 stream
.next_in
= ibuf
;
201 stream
.avail_in
= readlen
;
202 while ((stream
.avail_in
|| readlen
== 0) &&
203 (zret
== Z_OK
|| zret
== Z_BUF_ERROR
)) {
204 stream
.next_out
= obuf
;
205 stream
.avail_out
= sizeof(obuf
);
206 zret
= git_deflate(&stream
, readlen
? 0 : Z_FINISH
);
207 hashwrite(f
, obuf
, stream
.next_out
- obuf
);
208 olen
+= stream
.next_out
- obuf
;
211 die(_("deflate error (%d)"), zret
);
213 if (zret
!= Z_STREAM_END
)
214 die(_("deflate error (%d)"), zret
);
218 git_deflate_end(&stream
);
223 * we are going to reuse the existing object data as is. make
224 * sure it is not corrupt.
226 static int check_pack_inflate(struct packed_git
*p
,
227 struct pack_window
**w_curs
,
230 unsigned long expect
)
233 unsigned char fakebuf
[4096], *in
;
236 memset(&stream
, 0, sizeof(stream
));
237 git_inflate_init(&stream
);
239 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
241 stream
.next_out
= fakebuf
;
242 stream
.avail_out
= sizeof(fakebuf
);
243 st
= git_inflate(&stream
, Z_FINISH
);
244 offset
+= stream
.next_in
- in
;
245 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
246 git_inflate_end(&stream
);
247 return (st
== Z_STREAM_END
&&
248 stream
.total_out
== expect
&&
249 stream
.total_in
== len
) ? 0 : -1;
252 static void copy_pack_data(struct hashfile
*f
,
253 struct packed_git
*p
,
254 struct pack_window
**w_curs
,
262 in
= use_pack(p
, w_curs
, offset
, &avail
);
264 avail
= (unsigned long)len
;
265 hashwrite(f
, in
, avail
);
271 /* Return 0 if we will bust the pack-size limit */
272 static unsigned long write_no_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
273 unsigned long limit
, int usable_delta
)
275 unsigned long size
, datalen
;
276 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
277 dheader
[MAX_PACK_OBJECT_HEADER
];
279 enum object_type type
;
281 struct git_istream
*st
= NULL
;
284 if (oe_type(entry
) == OBJ_BLOB
&&
285 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
) &&
286 (st
= open_istream(&entry
->idx
.oid
, &type
, &size
, NULL
)) != NULL
)
289 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
291 die(_("unable to read %s"),
292 oid_to_hex(&entry
->idx
.oid
));
295 * make sure no cached delta data remains from a
296 * previous attempt before a pack split occurred.
298 FREE_AND_NULL(entry
->delta_data
);
299 entry
->z_delta_size
= 0;
300 } else if (entry
->delta_data
) {
301 size
= DELTA_SIZE(entry
);
302 buf
= entry
->delta_data
;
303 entry
->delta_data
= NULL
;
304 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
305 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
307 buf
= get_delta(entry
);
308 size
= DELTA_SIZE(entry
);
309 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
310 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
313 if (st
) /* large blob case, just assume we don't compress well */
315 else if (entry
->z_delta_size
)
316 datalen
= entry
->z_delta_size
;
318 datalen
= do_compress(&buf
, size
);
321 * The object header is a byte of 'type' followed by zero or
322 * more bytes of length.
324 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
327 if (type
== OBJ_OFS_DELTA
) {
329 * Deltas with relative base contain an additional
330 * encoding of the relative offset for the delta
331 * base from this object's position in the pack.
333 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
334 unsigned pos
= sizeof(dheader
) - 1;
335 dheader
[pos
] = ofs
& 127;
337 dheader
[--pos
] = 128 | (--ofs
& 127);
338 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
344 hashwrite(f
, header
, hdrlen
);
345 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
346 hdrlen
+= sizeof(dheader
) - pos
;
347 } else if (type
== OBJ_REF_DELTA
) {
349 * Deltas with a base reference contain
350 * an additional 20 bytes for the base sha1.
352 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
358 hashwrite(f
, header
, hdrlen
);
359 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, 20);
362 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
368 hashwrite(f
, header
, hdrlen
);
371 datalen
= write_large_blob_data(st
, f
, &entry
->idx
.oid
);
374 hashwrite(f
, buf
, datalen
);
378 return hdrlen
+ datalen
;
381 /* Return 0 if we will bust the pack-size limit */
382 static off_t
write_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
383 unsigned long limit
, int usable_delta
)
385 struct packed_git
*p
= IN_PACK(entry
);
386 struct pack_window
*w_curs
= NULL
;
387 struct revindex_entry
*revidx
;
389 enum object_type type
= oe_type(entry
);
391 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
392 dheader
[MAX_PACK_OBJECT_HEADER
];
394 unsigned long entry_size
= SIZE(entry
);
397 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
398 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
399 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
402 offset
= entry
->in_pack_offset
;
403 revidx
= find_pack_revindex(p
, offset
);
404 datalen
= revidx
[1].offset
- offset
;
405 if (!pack_to_stdout
&& p
->index_version
> 1 &&
406 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
)) {
407 error("bad packed object CRC for %s",
408 oid_to_hex(&entry
->idx
.oid
));
410 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
413 offset
+= entry
->in_pack_header_size
;
414 datalen
-= entry
->in_pack_header_size
;
416 if (!pack_to_stdout
&& p
->index_version
== 1 &&
417 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry_size
)) {
418 error("corrupt packed object for %s",
419 oid_to_hex(&entry
->idx
.oid
));
421 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
424 if (type
== OBJ_OFS_DELTA
) {
425 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
426 unsigned pos
= sizeof(dheader
) - 1;
427 dheader
[pos
] = ofs
& 127;
429 dheader
[--pos
] = 128 | (--ofs
& 127);
430 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
434 hashwrite(f
, header
, hdrlen
);
435 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
436 hdrlen
+= sizeof(dheader
) - pos
;
438 } else if (type
== OBJ_REF_DELTA
) {
439 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
443 hashwrite(f
, header
, hdrlen
);
444 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, 20);
448 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
452 hashwrite(f
, header
, hdrlen
);
454 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
457 return hdrlen
+ datalen
;
460 /* Return 0 if we will bust the pack-size limit */
461 static off_t
write_object(struct hashfile
*f
,
462 struct object_entry
*entry
,
467 int usable_delta
, to_reuse
;
472 /* apply size limit if limited packsize and not first object */
473 if (!pack_size_limit
|| !nr_written
)
475 else if (pack_size_limit
<= write_offset
)
477 * the earlier object did not fit the limit; avoid
478 * mistaking this with unlimited (i.e. limit = 0).
482 limit
= pack_size_limit
- write_offset
;
485 usable_delta
= 0; /* no delta */
486 else if (!pack_size_limit
)
487 usable_delta
= 1; /* unlimited packfile */
488 else if (DELTA(entry
)->idx
.offset
== (off_t
)-1)
489 usable_delta
= 0; /* base was written to another pack */
490 else if (DELTA(entry
)->idx
.offset
)
491 usable_delta
= 1; /* base already exists in this pack */
493 usable_delta
= 0; /* base could end up in another pack */
496 to_reuse
= 0; /* explicit */
497 else if (!IN_PACK(entry
))
498 to_reuse
= 0; /* can't reuse what we don't have */
499 else if (oe_type(entry
) == OBJ_REF_DELTA
||
500 oe_type(entry
) == OBJ_OFS_DELTA
)
501 /* check_object() decided it for us ... */
502 to_reuse
= usable_delta
;
503 /* ... but pack split may override that */
504 else if (oe_type(entry
) != entry
->in_pack_type
)
505 to_reuse
= 0; /* pack has delta which is unusable */
506 else if (DELTA(entry
))
507 to_reuse
= 0; /* we want to pack afresh */
509 to_reuse
= 1; /* we have it in-pack undeltified,
510 * and we do not need to deltify it.
514 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
516 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
524 entry
->idx
.crc32
= crc32_end(f
);
528 enum write_one_status
{
529 WRITE_ONE_SKIP
= -1, /* already written */
530 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
531 WRITE_ONE_WRITTEN
= 1, /* normal */
532 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
535 static enum write_one_status
write_one(struct hashfile
*f
,
536 struct object_entry
*e
,
543 * we set offset to 1 (which is an impossible value) to mark
544 * the fact that this object is involved in "write its base
545 * first before writing a deltified object" recursion.
547 recursing
= (e
->idx
.offset
== 1);
549 warning("recursive delta detected for object %s",
550 oid_to_hex(&e
->idx
.oid
));
551 return WRITE_ONE_RECURSIVE
;
552 } else if (e
->idx
.offset
|| e
->preferred_base
) {
553 /* offset is non zero if object is written already. */
554 return WRITE_ONE_SKIP
;
557 /* if we are deltified, write out base object first. */
559 e
->idx
.offset
= 1; /* now recurse */
560 switch (write_one(f
, DELTA(e
), offset
)) {
561 case WRITE_ONE_RECURSIVE
:
562 /* we cannot depend on this one */
567 case WRITE_ONE_BREAK
:
568 e
->idx
.offset
= recursing
;
569 return WRITE_ONE_BREAK
;
573 e
->idx
.offset
= *offset
;
574 size
= write_object(f
, e
, *offset
);
576 e
->idx
.offset
= recursing
;
577 return WRITE_ONE_BREAK
;
579 written_list
[nr_written
++] = &e
->idx
;
581 /* make sure off_t is sufficiently large not to wrap */
582 if (signed_add_overflows(*offset
, size
))
583 die("pack too large for current definition of off_t");
585 return WRITE_ONE_WRITTEN
;
588 static int mark_tagged(const char *path
, const struct object_id
*oid
, int flag
,
591 struct object_id peeled
;
592 struct object_entry
*entry
= packlist_find(&to_pack
, oid
->hash
, NULL
);
596 if (!peel_ref(path
, &peeled
)) {
597 entry
= packlist_find(&to_pack
, peeled
.hash
, NULL
);
604 static inline void add_to_write_order(struct object_entry
**wo
,
606 struct object_entry
*e
)
614 static void add_descendants_to_write_order(struct object_entry
**wo
,
616 struct object_entry
*e
)
618 int add_to_order
= 1;
621 struct object_entry
*s
;
622 /* add this node... */
623 add_to_write_order(wo
, endp
, e
);
624 /* all its siblings... */
625 for (s
= DELTA_SIBLING(e
); s
; s
= DELTA_SIBLING(s
)) {
626 add_to_write_order(wo
, endp
, s
);
629 /* drop down a level to add left subtree nodes if possible */
630 if (DELTA_CHILD(e
)) {
635 /* our sibling might have some children, it is next */
636 if (DELTA_SIBLING(e
)) {
637 e
= DELTA_SIBLING(e
);
640 /* go back to our parent node */
642 while (e
&& !DELTA_SIBLING(e
)) {
643 /* we're on the right side of a subtree, keep
644 * going up until we can go right again */
648 /* done- we hit our original root node */
651 /* pass it off to sibling at this level */
652 e
= DELTA_SIBLING(e
);
657 static void add_family_to_write_order(struct object_entry
**wo
,
659 struct object_entry
*e
)
661 struct object_entry
*root
;
663 for (root
= e
; DELTA(root
); root
= DELTA(root
))
665 add_descendants_to_write_order(wo
, endp
, root
);
668 static struct object_entry
**compute_write_order(void)
670 unsigned int i
, wo_end
, last_untagged
;
672 struct object_entry
**wo
;
673 struct object_entry
*objects
= to_pack
.objects
;
675 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
676 objects
[i
].tagged
= 0;
677 objects
[i
].filled
= 0;
678 SET_DELTA_CHILD(&objects
[i
], NULL
);
679 SET_DELTA_SIBLING(&objects
[i
], NULL
);
683 * Fully connect delta_child/delta_sibling network.
684 * Make sure delta_sibling is sorted in the original
687 for (i
= to_pack
.nr_objects
; i
> 0;) {
688 struct object_entry
*e
= &objects
[--i
];
691 /* Mark me as the first child */
692 e
->delta_sibling_idx
= DELTA(e
)->delta_child_idx
;
693 SET_DELTA_CHILD(DELTA(e
), e
);
697 * Mark objects that are at the tip of tags.
699 for_each_tag_ref(mark_tagged
, NULL
);
702 * Give the objects in the original recency order until
703 * we see a tagged tip.
705 ALLOC_ARRAY(wo
, to_pack
.nr_objects
);
706 for (i
= wo_end
= 0; i
< to_pack
.nr_objects
; i
++) {
707 if (objects
[i
].tagged
)
709 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
714 * Then fill all the tagged tips.
716 for (; i
< to_pack
.nr_objects
; i
++) {
717 if (objects
[i
].tagged
)
718 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
722 * And then all remaining commits and tags.
724 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
725 if (oe_type(&objects
[i
]) != OBJ_COMMIT
&&
726 oe_type(&objects
[i
]) != OBJ_TAG
)
728 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
732 * And then all the trees.
734 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
735 if (oe_type(&objects
[i
]) != OBJ_TREE
)
737 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
741 * Finally all the rest in really tight order
743 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
744 if (!objects
[i
].filled
)
745 add_family_to_write_order(wo
, &wo_end
, &objects
[i
]);
748 if (wo_end
!= to_pack
.nr_objects
)
749 die("ordered %u objects, expected %"PRIu32
, wo_end
, to_pack
.nr_objects
);
754 static off_t
write_reused_pack(struct hashfile
*f
)
756 unsigned char buffer
[8192];
757 off_t to_write
, total
;
760 if (!is_pack_valid(reuse_packfile
))
761 die("packfile is invalid: %s", reuse_packfile
->pack_name
);
763 fd
= git_open(reuse_packfile
->pack_name
);
765 die_errno("unable to open packfile for reuse: %s",
766 reuse_packfile
->pack_name
);
768 if (lseek(fd
, sizeof(struct pack_header
), SEEK_SET
) == -1)
769 die_errno("unable to seek in reused packfile");
771 if (reuse_packfile_offset
< 0)
772 reuse_packfile_offset
= reuse_packfile
->pack_size
- 20;
774 total
= to_write
= reuse_packfile_offset
- sizeof(struct pack_header
);
777 int read_pack
= xread(fd
, buffer
, sizeof(buffer
));
780 die_errno("unable to read from reused packfile");
782 if (read_pack
> to_write
)
783 read_pack
= to_write
;
785 hashwrite(f
, buffer
, read_pack
);
786 to_write
-= read_pack
;
789 * We don't know the actual number of objects written,
790 * only how many bytes written, how many bytes total, and
791 * how many objects total. So we can fake it by pretending all
792 * objects we are writing are the same size. This gives us a
793 * smooth progress meter, and at the end it matches the true
796 written
= reuse_packfile_objects
*
797 (((double)(total
- to_write
)) / total
);
798 display_progress(progress_state
, written
);
802 written
= reuse_packfile_objects
;
803 display_progress(progress_state
, written
);
804 return reuse_packfile_offset
- sizeof(struct pack_header
);
807 static const char no_split_warning
[] = N_(
808 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
811 static void write_pack_file(void)
816 uint32_t nr_remaining
= nr_result
;
817 time_t last_mtime
= 0;
818 struct object_entry
**write_order
;
820 if (progress
> pack_to_stdout
)
821 progress_state
= start_progress(_("Writing objects"), nr_result
);
822 ALLOC_ARRAY(written_list
, to_pack
.nr_objects
);
823 write_order
= compute_write_order();
826 struct object_id oid
;
827 char *pack_tmp_name
= NULL
;
830 f
= hashfd_throughput(1, "<stdout>", progress_state
);
832 f
= create_tmp_packfile(&pack_tmp_name
);
834 offset
= write_pack_header(f
, nr_remaining
);
836 if (reuse_packfile
) {
838 assert(pack_to_stdout
);
840 packfile_size
= write_reused_pack(f
);
841 offset
+= packfile_size
;
845 for (; i
< to_pack
.nr_objects
; i
++) {
846 struct object_entry
*e
= write_order
[i
];
847 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
849 display_progress(progress_state
, written
);
853 * Did we write the wrong # entries in the header?
854 * If so, rewrite it like in fast-import
856 if (pack_to_stdout
) {
857 finalize_hashfile(f
, oid
.hash
, CSUM_HASH_IN_STREAM
| CSUM_CLOSE
);
858 } else if (nr_written
== nr_remaining
) {
859 finalize_hashfile(f
, oid
.hash
, CSUM_HASH_IN_STREAM
| CSUM_FSYNC
| CSUM_CLOSE
);
861 int fd
= finalize_hashfile(f
, oid
.hash
, 0);
862 fixup_pack_header_footer(fd
, oid
.hash
, pack_tmp_name
,
863 nr_written
, oid
.hash
, offset
);
865 if (write_bitmap_index
) {
866 warning(_(no_split_warning
));
867 write_bitmap_index
= 0;
871 if (!pack_to_stdout
) {
873 struct strbuf tmpname
= STRBUF_INIT
;
876 * Packs are runtime accessed in their mtime
877 * order since newer packs are more likely to contain
878 * younger objects. So if we are creating multiple
879 * packs then we should modify the mtime of later ones
880 * to preserve this property.
882 if (stat(pack_tmp_name
, &st
) < 0) {
883 warning_errno("failed to stat %s", pack_tmp_name
);
884 } else if (!last_mtime
) {
885 last_mtime
= st
.st_mtime
;
888 utb
.actime
= st
.st_atime
;
889 utb
.modtime
= --last_mtime
;
890 if (utime(pack_tmp_name
, &utb
) < 0)
891 warning_errno("failed utime() on %s", pack_tmp_name
);
894 strbuf_addf(&tmpname
, "%s-", base_name
);
896 if (write_bitmap_index
) {
897 bitmap_writer_set_checksum(oid
.hash
);
898 bitmap_writer_build_type_index(
899 &to_pack
, written_list
, nr_written
);
902 finish_tmp_packfile(&tmpname
, pack_tmp_name
,
903 written_list
, nr_written
,
904 &pack_idx_opts
, oid
.hash
);
906 if (write_bitmap_index
) {
907 strbuf_addf(&tmpname
, "%s.bitmap", oid_to_hex(&oid
));
909 stop_progress(&progress_state
);
911 bitmap_writer_show_progress(progress
);
912 bitmap_writer_reuse_bitmaps(&to_pack
);
913 bitmap_writer_select_commits(indexed_commits
, indexed_commits_nr
, -1);
914 bitmap_writer_build(&to_pack
);
915 bitmap_writer_finish(written_list
, nr_written
,
916 tmpname
.buf
, write_bitmap_options
);
917 write_bitmap_index
= 0;
920 strbuf_release(&tmpname
);
922 puts(oid_to_hex(&oid
));
925 /* mark written objects as written to previous pack */
926 for (j
= 0; j
< nr_written
; j
++) {
927 written_list
[j
]->offset
= (off_t
)-1;
929 nr_remaining
-= nr_written
;
930 } while (nr_remaining
&& i
< to_pack
.nr_objects
);
934 stop_progress(&progress_state
);
935 if (written
!= nr_result
)
936 die("wrote %"PRIu32
" objects while expecting %"PRIu32
,
940 static int no_try_delta(const char *path
)
942 static struct attr_check
*check
;
945 check
= attr_check_initl("delta", NULL
);
946 if (git_check_attr(path
, check
))
948 if (ATTR_FALSE(check
->items
[0].value
))
954 * When adding an object, check whether we have already added it
955 * to our packing list. If so, we can skip. However, if we are
956 * being asked to excludei t, but the previous mention was to include
957 * it, make sure to adjust its flags and tweak our numbers accordingly.
959 * As an optimization, we pass out the index position where we would have
960 * found the item, since that saves us from having to look it up again a
961 * few lines later when we want to add the new entry.
963 static int have_duplicate_entry(const struct object_id
*oid
,
967 struct object_entry
*entry
;
969 entry
= packlist_find(&to_pack
, oid
->hash
, index_pos
);
974 if (!entry
->preferred_base
)
976 entry
->preferred_base
= 1;
982 static int want_found_object(int exclude
, struct packed_git
*p
)
990 * When asked to do --local (do not include an object that appears in a
991 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
992 * an object that appears in a pack marked with .keep), finding a pack
993 * that matches the criteria is sufficient for us to decide to omit it.
994 * However, even if this pack does not satisfy the criteria, we need to
995 * make sure no copy of this object appears in _any_ pack that makes us
996 * to omit the object, so we need to check all the packs.
998 * We can however first check whether these options can possible matter;
999 * if they do not matter we know we want the object in generated pack.
1000 * Otherwise, we signal "-1" at the end to tell the caller that we do
1001 * not know either way, and it needs to check more packs.
1003 if (!ignore_packed_keep_on_disk
&&
1004 !ignore_packed_keep_in_core
&&
1005 (!local
|| !have_non_local_packs
))
1008 if (local
&& !p
->pack_local
)
1010 if (p
->pack_local
&&
1011 ((ignore_packed_keep_on_disk
&& p
->pack_keep
) ||
1012 (ignore_packed_keep_in_core
&& p
->pack_keep_in_core
)))
1015 /* we don't know yet; keep looking for more packs */
1020 * Check whether we want the object in the pack (e.g., we do not want
1021 * objects found in non-local stores if the "--local" option was used).
1023 * If the caller already knows an existing pack it wants to take the object
1024 * from, that is passed in *found_pack and *found_offset; otherwise this
1025 * function finds if there is any pack that has the object and returns the pack
1026 * and its offset in these variables.
1028 static int want_object_in_pack(const struct object_id
*oid
,
1030 struct packed_git
**found_pack
,
1031 off_t
*found_offset
)
1034 struct list_head
*pos
;
1036 if (!exclude
&& local
&& has_loose_object_nonlocal(oid
->hash
))
1040 * If we already know the pack object lives in, start checks from that
1041 * pack - in the usual case when neither --local was given nor .keep files
1042 * are present we will determine the answer right now.
1045 want
= want_found_object(exclude
, *found_pack
);
1049 list_for_each(pos
, get_packed_git_mru(the_repository
)) {
1050 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1053 if (p
== *found_pack
)
1054 offset
= *found_offset
;
1056 offset
= find_pack_entry_one(oid
->hash
, p
);
1060 if (!is_pack_valid(p
))
1062 *found_offset
= offset
;
1065 want
= want_found_object(exclude
, p
);
1066 if (!exclude
&& want
> 0)
1068 get_packed_git_mru(the_repository
));
1077 static void create_object_entry(const struct object_id
*oid
,
1078 enum object_type type
,
1083 struct packed_git
*found_pack
,
1086 struct object_entry
*entry
;
1088 entry
= packlist_alloc(&to_pack
, oid
->hash
, index_pos
);
1090 oe_set_type(entry
, type
);
1092 entry
->preferred_base
= 1;
1096 oe_set_in_pack(&to_pack
, entry
, found_pack
);
1097 entry
->in_pack_offset
= found_offset
;
1100 entry
->no_try_delta
= no_try_delta
;
1103 static const char no_closure_warning
[] = N_(
1104 "disabling bitmap writing, as some objects are not being packed"
1107 static int add_object_entry(const struct object_id
*oid
, enum object_type type
,
1108 const char *name
, int exclude
)
1110 struct packed_git
*found_pack
= NULL
;
1111 off_t found_offset
= 0;
1114 display_progress(progress_state
, ++nr_seen
);
1116 if (have_duplicate_entry(oid
, exclude
, &index_pos
))
1119 if (!want_object_in_pack(oid
, exclude
, &found_pack
, &found_offset
)) {
1120 /* The pack is missing an object, so it will not have closure */
1121 if (write_bitmap_index
) {
1122 warning(_(no_closure_warning
));
1123 write_bitmap_index
= 0;
1128 create_object_entry(oid
, type
, pack_name_hash(name
),
1129 exclude
, name
&& no_try_delta(name
),
1130 index_pos
, found_pack
, found_offset
);
1134 static int add_object_entry_from_bitmap(const struct object_id
*oid
,
1135 enum object_type type
,
1136 int flags
, uint32_t name_hash
,
1137 struct packed_git
*pack
, off_t offset
)
1141 display_progress(progress_state
, ++nr_seen
);
1143 if (have_duplicate_entry(oid
, 0, &index_pos
))
1146 if (!want_object_in_pack(oid
, 0, &pack
, &offset
))
1149 create_object_entry(oid
, type
, name_hash
, 0, 0, index_pos
, pack
, offset
);
1153 struct pbase_tree_cache
{
1154 struct object_id oid
;
1158 unsigned long tree_size
;
1161 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1162 static int pbase_tree_cache_ix(const struct object_id
*oid
)
1164 return oid
->hash
[0] % ARRAY_SIZE(pbase_tree_cache
);
1166 static int pbase_tree_cache_ix_incr(int ix
)
1168 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1171 static struct pbase_tree
{
1172 struct pbase_tree
*next
;
1173 /* This is a phony "cache" entry; we are not
1174 * going to evict it or find it through _get()
1175 * mechanism -- this is for the toplevel node that
1176 * would almost always change with any commit.
1178 struct pbase_tree_cache pcache
;
1181 static struct pbase_tree_cache
*pbase_tree_get(const struct object_id
*oid
)
1183 struct pbase_tree_cache
*ent
, *nent
;
1186 enum object_type type
;
1188 int my_ix
= pbase_tree_cache_ix(oid
);
1189 int available_ix
= -1;
1191 /* pbase-tree-cache acts as a limited hashtable.
1192 * your object will be found at your index or within a few
1193 * slots after that slot if it is cached.
1195 for (neigh
= 0; neigh
< 8; neigh
++) {
1196 ent
= pbase_tree_cache
[my_ix
];
1197 if (ent
&& !oidcmp(&ent
->oid
, oid
)) {
1201 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1202 ((0 <= available_ix
) &&
1203 (!ent
&& pbase_tree_cache
[available_ix
])))
1204 available_ix
= my_ix
;
1207 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1210 /* Did not find one. Either we got a bogus request or
1211 * we need to read and perhaps cache.
1213 data
= read_object_file(oid
, &type
, &size
);
1216 if (type
!= OBJ_TREE
) {
1221 /* We need to either cache or return a throwaway copy */
1223 if (available_ix
< 0)
1226 ent
= pbase_tree_cache
[available_ix
];
1227 my_ix
= available_ix
;
1231 nent
= xmalloc(sizeof(*nent
));
1232 nent
->temporary
= (available_ix
< 0);
1235 /* evict and reuse */
1236 free(ent
->tree_data
);
1239 oidcpy(&nent
->oid
, oid
);
1240 nent
->tree_data
= data
;
1241 nent
->tree_size
= size
;
1243 if (!nent
->temporary
)
1244 pbase_tree_cache
[my_ix
] = nent
;
1248 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1250 if (!cache
->temporary
) {
1254 free(cache
->tree_data
);
1258 static int name_cmp_len(const char *name
)
1261 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
1266 static void add_pbase_object(struct tree_desc
*tree
,
1269 const char *fullname
)
1271 struct name_entry entry
;
1274 while (tree_entry(tree
,&entry
)) {
1275 if (S_ISGITLINK(entry
.mode
))
1277 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1278 memcmp(name
, entry
.path
, cmplen
);
1283 if (name
[cmplen
] != '/') {
1284 add_object_entry(entry
.oid
,
1285 object_type(entry
.mode
),
1289 if (S_ISDIR(entry
.mode
)) {
1290 struct tree_desc sub
;
1291 struct pbase_tree_cache
*tree
;
1292 const char *down
= name
+cmplen
+1;
1293 int downlen
= name_cmp_len(down
);
1295 tree
= pbase_tree_get(entry
.oid
);
1298 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1300 add_pbase_object(&sub
, down
, downlen
, fullname
);
1301 pbase_tree_put(tree
);
1306 static unsigned *done_pbase_paths
;
1307 static int done_pbase_paths_num
;
1308 static int done_pbase_paths_alloc
;
1309 static int done_pbase_path_pos(unsigned hash
)
1312 int hi
= done_pbase_paths_num
;
1314 int mi
= lo
+ (hi
- lo
) / 2;
1315 if (done_pbase_paths
[mi
] == hash
)
1317 if (done_pbase_paths
[mi
] < hash
)
1325 static int check_pbase_path(unsigned hash
)
1327 int pos
= done_pbase_path_pos(hash
);
1331 ALLOC_GROW(done_pbase_paths
,
1332 done_pbase_paths_num
+ 1,
1333 done_pbase_paths_alloc
);
1334 done_pbase_paths_num
++;
1335 if (pos
< done_pbase_paths_num
)
1336 MOVE_ARRAY(done_pbase_paths
+ pos
+ 1, done_pbase_paths
+ pos
,
1337 done_pbase_paths_num
- pos
- 1);
1338 done_pbase_paths
[pos
] = hash
;
1342 static void add_preferred_base_object(const char *name
)
1344 struct pbase_tree
*it
;
1346 unsigned hash
= pack_name_hash(name
);
1348 if (!num_preferred_base
|| check_pbase_path(hash
))
1351 cmplen
= name_cmp_len(name
);
1352 for (it
= pbase_tree
; it
; it
= it
->next
) {
1354 add_object_entry(&it
->pcache
.oid
, OBJ_TREE
, NULL
, 1);
1357 struct tree_desc tree
;
1358 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1359 add_pbase_object(&tree
, name
, cmplen
, name
);
1364 static void add_preferred_base(struct object_id
*oid
)
1366 struct pbase_tree
*it
;
1369 struct object_id tree_oid
;
1371 if (window
<= num_preferred_base
++)
1374 data
= read_object_with_reference(oid
, tree_type
, &size
, &tree_oid
);
1378 for (it
= pbase_tree
; it
; it
= it
->next
) {
1379 if (!oidcmp(&it
->pcache
.oid
, &tree_oid
)) {
1385 it
= xcalloc(1, sizeof(*it
));
1386 it
->next
= pbase_tree
;
1389 oidcpy(&it
->pcache
.oid
, &tree_oid
);
1390 it
->pcache
.tree_data
= data
;
1391 it
->pcache
.tree_size
= size
;
1394 static void cleanup_preferred_base(void)
1396 struct pbase_tree
*it
;
1402 struct pbase_tree
*tmp
= it
;
1404 free(tmp
->pcache
.tree_data
);
1408 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1409 if (!pbase_tree_cache
[i
])
1411 free(pbase_tree_cache
[i
]->tree_data
);
1412 FREE_AND_NULL(pbase_tree_cache
[i
]);
1415 FREE_AND_NULL(done_pbase_paths
);
1416 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1419 static void check_object(struct object_entry
*entry
)
1421 unsigned long canonical_size
;
1423 if (IN_PACK(entry
)) {
1424 struct packed_git
*p
= IN_PACK(entry
);
1425 struct pack_window
*w_curs
= NULL
;
1426 const unsigned char *base_ref
= NULL
;
1427 struct object_entry
*base_entry
;
1428 unsigned long used
, used_0
;
1429 unsigned long avail
;
1431 unsigned char *buf
, c
;
1432 enum object_type type
;
1433 unsigned long in_pack_size
;
1435 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1438 * We want in_pack_type even if we do not reuse delta
1439 * since non-delta representations could still be reused.
1441 used
= unpack_object_header_buffer(buf
, avail
,
1448 BUG("invalid type %d", type
);
1449 entry
->in_pack_type
= type
;
1452 * Determine if this is a delta and if so whether we can
1453 * reuse it or not. Otherwise let's find out as cheaply as
1454 * possible what the actual type and size for this object is.
1456 switch (entry
->in_pack_type
) {
1458 /* Not a delta hence we've already got all we need. */
1459 oe_set_type(entry
, entry
->in_pack_type
);
1460 SET_SIZE(entry
, in_pack_size
);
1461 entry
->in_pack_header_size
= used
;
1462 if (oe_type(entry
) < OBJ_COMMIT
|| oe_type(entry
) > OBJ_BLOB
)
1464 unuse_pack(&w_curs
);
1467 if (reuse_delta
&& !entry
->preferred_base
)
1468 base_ref
= use_pack(p
, &w_curs
,
1469 entry
->in_pack_offset
+ used
, NULL
);
1470 entry
->in_pack_header_size
= used
+ 20;
1473 buf
= use_pack(p
, &w_curs
,
1474 entry
->in_pack_offset
+ used
, NULL
);
1480 if (!ofs
|| MSB(ofs
, 7)) {
1481 error("delta base offset overflow in pack for %s",
1482 oid_to_hex(&entry
->idx
.oid
));
1486 ofs
= (ofs
<< 7) + (c
& 127);
1488 ofs
= entry
->in_pack_offset
- ofs
;
1489 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1490 error("delta base offset out of bound for %s",
1491 oid_to_hex(&entry
->idx
.oid
));
1494 if (reuse_delta
&& !entry
->preferred_base
) {
1495 struct revindex_entry
*revidx
;
1496 revidx
= find_pack_revindex(p
, ofs
);
1499 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1501 entry
->in_pack_header_size
= used
+ used_0
;
1505 if (base_ref
&& (base_entry
= packlist_find(&to_pack
, base_ref
, NULL
))) {
1507 * If base_ref was set above that means we wish to
1508 * reuse delta data, and we even found that base
1509 * in the list of objects we want to pack. Goodie!
1511 * Depth value does not matter - find_deltas() will
1512 * never consider reused delta as the base object to
1513 * deltify other objects against, in order to avoid
1516 oe_set_type(entry
, entry
->in_pack_type
);
1517 SET_SIZE(entry
, in_pack_size
); /* delta size */
1518 SET_DELTA(entry
, base_entry
);
1519 SET_DELTA_SIZE(entry
, in_pack_size
);
1520 entry
->delta_sibling_idx
= base_entry
->delta_child_idx
;
1521 SET_DELTA_CHILD(base_entry
, entry
);
1522 unuse_pack(&w_curs
);
1526 if (oe_type(entry
)) {
1530 * This must be a delta and we already know what the
1531 * final object type is. Let's extract the actual
1532 * object size from the delta header.
1534 delta_pos
= entry
->in_pack_offset
+ entry
->in_pack_header_size
;
1535 canonical_size
= get_size_from_delta(p
, &w_curs
, delta_pos
);
1536 if (canonical_size
== 0)
1538 SET_SIZE(entry
, canonical_size
);
1539 unuse_pack(&w_curs
);
1544 * No choice but to fall back to the recursive delta walk
1545 * with sha1_object_info() to find about the object type
1549 unuse_pack(&w_curs
);
1553 oid_object_info(the_repository
, &entry
->idx
.oid
, &canonical_size
));
1554 if (entry
->type_valid
) {
1555 SET_SIZE(entry
, canonical_size
);
1558 * Bad object type is checked in prepare_pack(). This is
1559 * to permit a missing preferred base object to be ignored
1560 * as a preferred base. Doing so can result in a larger
1561 * pack file, but the transfer will still take place.
1566 static int pack_offset_sort(const void *_a
, const void *_b
)
1568 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1569 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1570 const struct packed_git
*a_in_pack
= IN_PACK(a
);
1571 const struct packed_git
*b_in_pack
= IN_PACK(b
);
1573 /* avoid filesystem trashing with loose objects */
1574 if (!a_in_pack
&& !b_in_pack
)
1575 return oidcmp(&a
->idx
.oid
, &b
->idx
.oid
);
1577 if (a_in_pack
< b_in_pack
)
1579 if (a_in_pack
> b_in_pack
)
1581 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1582 (a
->in_pack_offset
> b
->in_pack_offset
);
1586 * Drop an on-disk delta we were planning to reuse. Naively, this would
1587 * just involve blanking out the "delta" field, but we have to deal
1588 * with some extra book-keeping:
1590 * 1. Removing ourselves from the delta_sibling linked list.
1592 * 2. Updating our size/type to the non-delta representation. These were
1593 * either not recorded initially (size) or overwritten with the delta type
1594 * (type) when check_object() decided to reuse the delta.
1596 * 3. Resetting our delta depth, as we are now a base object.
1598 static void drop_reused_delta(struct object_entry
*entry
)
1600 unsigned *idx
= &to_pack
.objects
[entry
->delta_idx
- 1].delta_child_idx
;
1601 struct object_info oi
= OBJECT_INFO_INIT
;
1602 enum object_type type
;
1606 struct object_entry
*oe
= &to_pack
.objects
[*idx
- 1];
1609 *idx
= oe
->delta_sibling_idx
;
1611 idx
= &oe
->delta_sibling_idx
;
1613 SET_DELTA(entry
, NULL
);
1618 if (packed_object_info(the_repository
, IN_PACK(entry
), entry
->in_pack_offset
, &oi
) < 0) {
1620 * We failed to get the info from this pack for some reason;
1621 * fall back to sha1_object_info, which may find another copy.
1622 * And if that fails, the error will be recorded in oe_type(entry)
1623 * and dealt with in prepare_pack().
1626 oid_object_info(the_repository
, &entry
->idx
.oid
, &size
));
1628 oe_set_type(entry
, type
);
1630 SET_SIZE(entry
, size
);
1634 * Follow the chain of deltas from this entry onward, throwing away any links
1635 * that cause us to hit a cycle (as determined by the DFS state flags in
1638 * We also detect too-long reused chains that would violate our --depth
1641 static void break_delta_chains(struct object_entry
*entry
)
1644 * The actual depth of each object we will write is stored as an int,
1645 * as it cannot exceed our int "depth" limit. But before we break
1646 * changes based no that limit, we may potentially go as deep as the
1647 * number of objects, which is elsewhere bounded to a uint32_t.
1649 uint32_t total_depth
;
1650 struct object_entry
*cur
, *next
;
1652 for (cur
= entry
, total_depth
= 0;
1654 cur
= DELTA(cur
), total_depth
++) {
1655 if (cur
->dfs_state
== DFS_DONE
) {
1657 * We've already seen this object and know it isn't
1658 * part of a cycle. We do need to append its depth
1661 total_depth
+= cur
->depth
;
1666 * We break cycles before looping, so an ACTIVE state (or any
1667 * other cruft which made its way into the state variable)
1670 if (cur
->dfs_state
!= DFS_NONE
)
1671 die("BUG: confusing delta dfs state in first pass: %d",
1675 * Now we know this is the first time we've seen the object. If
1676 * it's not a delta, we're done traversing, but we'll mark it
1677 * done to save time on future traversals.
1680 cur
->dfs_state
= DFS_DONE
;
1685 * Mark ourselves as active and see if the next step causes
1686 * us to cycle to another active object. It's important to do
1687 * this _before_ we loop, because it impacts where we make the
1688 * cut, and thus how our total_depth counter works.
1689 * E.g., We may see a partial loop like:
1691 * A -> B -> C -> D -> B
1693 * Cutting B->C breaks the cycle. But now the depth of A is
1694 * only 1, and our total_depth counter is at 3. The size of the
1695 * error is always one less than the size of the cycle we
1696 * broke. Commits C and D were "lost" from A's chain.
1698 * If we instead cut D->B, then the depth of A is correct at 3.
1699 * We keep all commits in the chain that we examined.
1701 cur
->dfs_state
= DFS_ACTIVE
;
1702 if (DELTA(cur
)->dfs_state
== DFS_ACTIVE
) {
1703 drop_reused_delta(cur
);
1704 cur
->dfs_state
= DFS_DONE
;
1710 * And now that we've gone all the way to the bottom of the chain, we
1711 * need to clear the active flags and set the depth fields as
1712 * appropriate. Unlike the loop above, which can quit when it drops a
1713 * delta, we need to keep going to look for more depth cuts. So we need
1714 * an extra "next" pointer to keep going after we reset cur->delta.
1716 for (cur
= entry
; cur
; cur
= next
) {
1720 * We should have a chain of zero or more ACTIVE states down to
1721 * a final DONE. We can quit after the DONE, because either it
1722 * has no bases, or we've already handled them in a previous
1725 if (cur
->dfs_state
== DFS_DONE
)
1727 else if (cur
->dfs_state
!= DFS_ACTIVE
)
1728 die("BUG: confusing delta dfs state in second pass: %d",
1732 * If the total_depth is more than depth, then we need to snip
1733 * the chain into two or more smaller chains that don't exceed
1734 * the maximum depth. Most of the resulting chains will contain
1735 * (depth + 1) entries (i.e., depth deltas plus one base), and
1736 * the last chain (i.e., the one containing entry) will contain
1737 * whatever entries are left over, namely
1738 * (total_depth % (depth + 1)) of them.
1740 * Since we are iterating towards decreasing depth, we need to
1741 * decrement total_depth as we go, and we need to write to the
1742 * entry what its final depth will be after all of the
1743 * snipping. Since we're snipping into chains of length (depth
1744 * + 1) entries, the final depth of an entry will be its
1745 * original depth modulo (depth + 1). Any time we encounter an
1746 * entry whose final depth is supposed to be zero, we snip it
1747 * from its delta base, thereby making it so.
1749 cur
->depth
= (total_depth
--) % (depth
+ 1);
1751 drop_reused_delta(cur
);
1753 cur
->dfs_state
= DFS_DONE
;
1757 static void get_object_details(void)
1760 struct object_entry
**sorted_by_offset
;
1763 progress_state
= start_progress(_("Counting objects"),
1764 to_pack
.nr_objects
);
1766 sorted_by_offset
= xcalloc(to_pack
.nr_objects
, sizeof(struct object_entry
*));
1767 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1768 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
1769 QSORT(sorted_by_offset
, to_pack
.nr_objects
, pack_offset_sort
);
1771 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
1772 struct object_entry
*entry
= sorted_by_offset
[i
];
1773 check_object(entry
);
1774 if (entry
->type_valid
&&
1775 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
))
1776 entry
->no_try_delta
= 1;
1777 display_progress(progress_state
, i
+ 1);
1779 stop_progress(&progress_state
);
1782 * This must happen in a second pass, since we rely on the delta
1783 * information for the whole list being completed.
1785 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1786 break_delta_chains(&to_pack
.objects
[i
]);
1788 free(sorted_by_offset
);
1792 * We search for deltas in a list sorted by type, by filename hash, and then
1793 * by size, so that we see progressively smaller and smaller files.
1794 * That's because we prefer deltas to be from the bigger file
1795 * to the smaller -- deletes are potentially cheaper, but perhaps
1796 * more importantly, the bigger file is likely the more recent
1797 * one. The deepest deltas are therefore the oldest objects which are
1798 * less susceptible to be accessed often.
1800 static int type_size_sort(const void *_a
, const void *_b
)
1802 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1803 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1804 enum object_type a_type
= oe_type(a
);
1805 enum object_type b_type
= oe_type(b
);
1806 unsigned long a_size
= SIZE(a
);
1807 unsigned long b_size
= SIZE(b
);
1809 if (a_type
> b_type
)
1811 if (a_type
< b_type
)
1813 if (a
->hash
> b
->hash
)
1815 if (a
->hash
< b
->hash
)
1817 if (a
->preferred_base
> b
->preferred_base
)
1819 if (a
->preferred_base
< b
->preferred_base
)
1821 if (a_size
> b_size
)
1823 if (a_size
< b_size
)
1825 return a
< b
? -1 : (a
> b
); /* newest first */
1829 struct object_entry
*entry
;
1831 struct delta_index
*index
;
1835 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1836 unsigned long delta_size
)
1838 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1841 if (delta_size
< cache_max_small_delta_size
)
1844 /* cache delta, if objects are large enough compared to delta size */
1845 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1853 static pthread_mutex_t read_mutex
;
1854 #define read_lock() pthread_mutex_lock(&read_mutex)
1855 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1857 static pthread_mutex_t cache_mutex
;
1858 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1859 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1861 static pthread_mutex_t progress_mutex
;
1862 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1863 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1867 #define read_lock() (void)0
1868 #define read_unlock() (void)0
1869 #define cache_lock() (void)0
1870 #define cache_unlock() (void)0
1871 #define progress_lock() (void)0
1872 #define progress_unlock() (void)0
1877 * Return the size of the object without doing any delta
1878 * reconstruction (so non-deltas are true object sizes, but deltas
1879 * return the size of the delta data).
1881 unsigned long oe_get_size_slow(struct packing_data
*pack
,
1882 const struct object_entry
*e
)
1884 struct packed_git
*p
;
1885 struct pack_window
*w_curs
;
1887 enum object_type type
;
1888 unsigned long used
, avail
, size
;
1890 if (e
->type_
!= OBJ_OFS_DELTA
&& e
->type_
!= OBJ_REF_DELTA
) {
1892 if (oid_object_info(the_repository
, &e
->idx
.oid
, &size
) < 0)
1893 die(_("unable to get size of %s"),
1894 oid_to_hex(&e
->idx
.oid
));
1899 p
= oe_in_pack(pack
, e
);
1901 BUG("when e->type is a delta, it must belong to a pack");
1905 buf
= use_pack(p
, &w_curs
, e
->in_pack_offset
, &avail
);
1906 used
= unpack_object_header_buffer(buf
, avail
, &type
, &size
);
1908 die(_("unable to parse object header of %s"),
1909 oid_to_hex(&e
->idx
.oid
));
1911 unuse_pack(&w_curs
);
1916 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1917 unsigned max_depth
, unsigned long *mem_usage
)
1919 struct object_entry
*trg_entry
= trg
->entry
;
1920 struct object_entry
*src_entry
= src
->entry
;
1921 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1923 enum object_type type
;
1926 /* Don't bother doing diffs between different types */
1927 if (oe_type(trg_entry
) != oe_type(src_entry
))
1931 * We do not bother to try a delta that we discarded on an
1932 * earlier try, but only when reusing delta data. Note that
1933 * src_entry that is marked as the preferred_base should always
1934 * be considered, as even if we produce a suboptimal delta against
1935 * it, we will still save the transfer cost, as we already know
1936 * the other side has it and we won't send src_entry at all.
1938 if (reuse_delta
&& IN_PACK(trg_entry
) &&
1939 IN_PACK(trg_entry
) == IN_PACK(src_entry
) &&
1940 !src_entry
->preferred_base
&&
1941 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1942 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1945 /* Let's not bust the allowed depth. */
1946 if (src
->depth
>= max_depth
)
1949 /* Now some size filtering heuristics. */
1950 trg_size
= SIZE(trg_entry
);
1951 if (!DELTA(trg_entry
)) {
1952 max_size
= trg_size
/2 - 20;
1955 max_size
= DELTA_SIZE(trg_entry
);
1956 ref_depth
= trg
->depth
;
1958 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
1959 (max_depth
- ref_depth
+ 1);
1962 src_size
= SIZE(src_entry
);
1963 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1964 if (sizediff
>= max_size
)
1966 if (trg_size
< src_size
/ 32)
1969 /* Load data if not already done */
1972 trg
->data
= read_object_file(&trg_entry
->idx
.oid
, &type
, &sz
);
1975 die("object %s cannot be read",
1976 oid_to_hex(&trg_entry
->idx
.oid
));
1978 die("object %s inconsistent object length (%lu vs %lu)",
1979 oid_to_hex(&trg_entry
->idx
.oid
), sz
,
1985 src
->data
= read_object_file(&src_entry
->idx
.oid
, &type
, &sz
);
1988 if (src_entry
->preferred_base
) {
1989 static int warned
= 0;
1991 warning("object %s cannot be read",
1992 oid_to_hex(&src_entry
->idx
.oid
));
1994 * Those objects are not included in the
1995 * resulting pack. Be resilient and ignore
1996 * them if they can't be read, in case the
1997 * pack could be created nevertheless.
2001 die("object %s cannot be read",
2002 oid_to_hex(&src_entry
->idx
.oid
));
2005 die("object %s inconsistent object length (%lu vs %lu)",
2006 oid_to_hex(&src_entry
->idx
.oid
), sz
,
2011 src
->index
= create_delta_index(src
->data
, src_size
);
2013 static int warned
= 0;
2015 warning("suboptimal pack - out of memory");
2018 *mem_usage
+= sizeof_delta_index(src
->index
);
2021 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
2024 if (delta_size
>= (1U << OE_DELTA_SIZE_BITS
)) {
2029 if (DELTA(trg_entry
)) {
2030 /* Prefer only shallower same-sized deltas. */
2031 if (delta_size
== DELTA_SIZE(trg_entry
) &&
2032 src
->depth
+ 1 >= trg
->depth
) {
2039 * Handle memory allocation outside of the cache
2040 * accounting lock. Compiler will optimize the strangeness
2041 * away when NO_PTHREADS is defined.
2043 free(trg_entry
->delta_data
);
2045 if (trg_entry
->delta_data
) {
2046 delta_cache_size
-= DELTA_SIZE(trg_entry
);
2047 trg_entry
->delta_data
= NULL
;
2049 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
2050 delta_cache_size
+= delta_size
;
2052 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
2058 SET_DELTA(trg_entry
, src_entry
);
2059 SET_DELTA_SIZE(trg_entry
, delta_size
);
2060 trg
->depth
= src
->depth
+ 1;
2065 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
2067 struct object_entry
*child
= DELTA_CHILD(me
);
2070 unsigned int c
= check_delta_limit(child
, n
+ 1);
2073 child
= DELTA_SIBLING(child
);
2078 static unsigned long free_unpacked(struct unpacked
*n
)
2080 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
2081 free_delta_index(n
->index
);
2084 freed_mem
+= SIZE(n
->entry
);
2085 FREE_AND_NULL(n
->data
);
2092 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
2093 int window
, int depth
, unsigned *processed
)
2095 uint32_t i
, idx
= 0, count
= 0;
2096 struct unpacked
*array
;
2097 unsigned long mem_usage
= 0;
2099 array
= xcalloc(window
, sizeof(struct unpacked
));
2102 struct object_entry
*entry
;
2103 struct unpacked
*n
= array
+ idx
;
2104 int j
, max_depth
, best_base
= -1;
2113 if (!entry
->preferred_base
) {
2115 display_progress(progress_state
, *processed
);
2119 mem_usage
-= free_unpacked(n
);
2122 while (window_memory_limit
&&
2123 mem_usage
> window_memory_limit
&&
2125 uint32_t tail
= (idx
+ window
- count
) % window
;
2126 mem_usage
-= free_unpacked(array
+ tail
);
2130 /* We do not compute delta to *create* objects we are not
2133 if (entry
->preferred_base
)
2137 * If the current object is at pack edge, take the depth the
2138 * objects that depend on the current object into account
2139 * otherwise they would become too deep.
2142 if (DELTA_CHILD(entry
)) {
2143 max_depth
-= check_delta_limit(entry
, 0);
2151 uint32_t other_idx
= idx
+ j
;
2153 if (other_idx
>= window
)
2154 other_idx
-= window
;
2155 m
= array
+ other_idx
;
2158 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
2162 best_base
= other_idx
;
2166 * If we decided to cache the delta data, then it is best
2167 * to compress it right away. First because we have to do
2168 * it anyway, and doing it here while we're threaded will
2169 * save a lot of time in the non threaded write phase,
2170 * as well as allow for caching more deltas within
2171 * the same cache size limit.
2173 * But only if not writing to stdout, since in that case
2174 * the network is most likely throttling writes anyway,
2175 * and therefore it is best to go to the write phase ASAP
2176 * instead, as we can afford spending more time compressing
2177 * between writes at that moment.
2179 if (entry
->delta_data
&& !pack_to_stdout
) {
2182 size
= do_compress(&entry
->delta_data
, DELTA_SIZE(entry
));
2183 if (size
< (1U << OE_Z_DELTA_BITS
)) {
2184 entry
->z_delta_size
= size
;
2186 delta_cache_size
-= DELTA_SIZE(entry
);
2187 delta_cache_size
+= entry
->z_delta_size
;
2190 FREE_AND_NULL(entry
->delta_data
);
2191 entry
->z_delta_size
= 0;
2195 /* if we made n a delta, and if n is already at max
2196 * depth, leaving it in the window is pointless. we
2197 * should evict it first.
2199 if (DELTA(entry
) && max_depth
<= n
->depth
)
2203 * Move the best delta base up in the window, after the
2204 * currently deltified object, to keep it longer. It will
2205 * be the first base object to be attempted next.
2208 struct unpacked swap
= array
[best_base
];
2209 int dist
= (window
+ idx
- best_base
) % window
;
2210 int dst
= best_base
;
2212 int src
= (dst
+ 1) % window
;
2213 array
[dst
] = array
[src
];
2221 if (count
+ 1 < window
)
2227 for (i
= 0; i
< window
; ++i
) {
2228 free_delta_index(array
[i
].index
);
2229 free(array
[i
].data
);
2236 static void try_to_free_from_threads(size_t size
)
2239 release_pack_memory(size
);
2243 static try_to_free_t old_try_to_free_routine
;
2246 * The main thread waits on the condition that (at least) one of the workers
2247 * has stopped working (which is indicated in the .working member of
2248 * struct thread_params).
2249 * When a work thread has completed its work, it sets .working to 0 and
2250 * signals the main thread and waits on the condition that .data_ready
2254 struct thread_params
{
2256 struct object_entry
**list
;
2263 pthread_mutex_t mutex
;
2264 pthread_cond_t cond
;
2265 unsigned *processed
;
2268 static pthread_cond_t progress_cond
;
2271 * Mutex and conditional variable can't be statically-initialized on Windows.
2273 static void init_threaded_search(void)
2275 init_recursive_mutex(&read_mutex
);
2276 pthread_mutex_init(&cache_mutex
, NULL
);
2277 pthread_mutex_init(&progress_mutex
, NULL
);
2278 pthread_cond_init(&progress_cond
, NULL
);
2279 old_try_to_free_routine
= set_try_to_free_routine(try_to_free_from_threads
);
2282 static void cleanup_threaded_search(void)
2284 set_try_to_free_routine(old_try_to_free_routine
);
2285 pthread_cond_destroy(&progress_cond
);
2286 pthread_mutex_destroy(&read_mutex
);
2287 pthread_mutex_destroy(&cache_mutex
);
2288 pthread_mutex_destroy(&progress_mutex
);
2291 static void *threaded_find_deltas(void *arg
)
2293 struct thread_params
*me
= arg
;
2296 while (me
->remaining
) {
2299 find_deltas(me
->list
, &me
->remaining
,
2300 me
->window
, me
->depth
, me
->processed
);
2304 pthread_cond_signal(&progress_cond
);
2308 * We must not set ->data_ready before we wait on the
2309 * condition because the main thread may have set it to 1
2310 * before we get here. In order to be sure that new
2311 * work is available if we see 1 in ->data_ready, it
2312 * was initialized to 0 before this thread was spawned
2313 * and we reset it to 0 right away.
2315 pthread_mutex_lock(&me
->mutex
);
2316 while (!me
->data_ready
)
2317 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2319 pthread_mutex_unlock(&me
->mutex
);
2324 /* leave ->working 1 so that this doesn't get more work assigned */
2328 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2329 int window
, int depth
, unsigned *processed
)
2331 struct thread_params
*p
;
2332 int i
, ret
, active_threads
= 0;
2334 init_threaded_search();
2336 if (delta_search_threads
<= 1) {
2337 find_deltas(list
, &list_size
, window
, depth
, processed
);
2338 cleanup_threaded_search();
2341 if (progress
> pack_to_stdout
)
2342 fprintf(stderr
, "Delta compression using up to %d threads.\n",
2343 delta_search_threads
);
2344 p
= xcalloc(delta_search_threads
, sizeof(*p
));
2346 /* Partition the work amongst work threads. */
2347 for (i
= 0; i
< delta_search_threads
; i
++) {
2348 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2350 /* don't use too small segments or no deltas will be found */
2351 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2354 p
[i
].window
= window
;
2356 p
[i
].processed
= processed
;
2358 p
[i
].data_ready
= 0;
2360 /* try to split chunks on "path" boundaries */
2361 while (sub_size
&& sub_size
< list_size
&&
2362 list
[sub_size
]->hash
&&
2363 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2367 p
[i
].list_size
= sub_size
;
2368 p
[i
].remaining
= sub_size
;
2371 list_size
-= sub_size
;
2374 /* Start work threads. */
2375 for (i
= 0; i
< delta_search_threads
; i
++) {
2376 if (!p
[i
].list_size
)
2378 pthread_mutex_init(&p
[i
].mutex
, NULL
);
2379 pthread_cond_init(&p
[i
].cond
, NULL
);
2380 ret
= pthread_create(&p
[i
].thread
, NULL
,
2381 threaded_find_deltas
, &p
[i
]);
2383 die("unable to create thread: %s", strerror(ret
));
2388 * Now let's wait for work completion. Each time a thread is done
2389 * with its work, we steal half of the remaining work from the
2390 * thread with the largest number of unprocessed objects and give
2391 * it to that newly idle thread. This ensure good load balancing
2392 * until the remaining object list segments are simply too short
2393 * to be worth splitting anymore.
2395 while (active_threads
) {
2396 struct thread_params
*target
= NULL
;
2397 struct thread_params
*victim
= NULL
;
2398 unsigned sub_size
= 0;
2402 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
2407 pthread_cond_wait(&progress_cond
, &progress_mutex
);
2410 for (i
= 0; i
< delta_search_threads
; i
++)
2411 if (p
[i
].remaining
> 2*window
&&
2412 (!victim
|| victim
->remaining
< p
[i
].remaining
))
2415 sub_size
= victim
->remaining
/ 2;
2416 list
= victim
->list
+ victim
->list_size
- sub_size
;
2417 while (sub_size
&& list
[0]->hash
&&
2418 list
[0]->hash
== list
[-1]->hash
) {
2424 * It is possible for some "paths" to have
2425 * so many objects that no hash boundary
2426 * might be found. Let's just steal the
2427 * exact half in that case.
2429 sub_size
= victim
->remaining
/ 2;
2432 target
->list
= list
;
2433 victim
->list_size
-= sub_size
;
2434 victim
->remaining
-= sub_size
;
2436 target
->list_size
= sub_size
;
2437 target
->remaining
= sub_size
;
2438 target
->working
= 1;
2441 pthread_mutex_lock(&target
->mutex
);
2442 target
->data_ready
= 1;
2443 pthread_cond_signal(&target
->cond
);
2444 pthread_mutex_unlock(&target
->mutex
);
2447 pthread_join(target
->thread
, NULL
);
2448 pthread_cond_destroy(&target
->cond
);
2449 pthread_mutex_destroy(&target
->mutex
);
2453 cleanup_threaded_search();
2458 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2461 static void add_tag_chain(const struct object_id
*oid
)
2466 * We catch duplicates already in add_object_entry(), but we'd
2467 * prefer to do this extra check to avoid having to parse the
2468 * tag at all if we already know that it's being packed (e.g., if
2469 * it was included via bitmaps, we would not have parsed it
2472 if (packlist_find(&to_pack
, oid
->hash
, NULL
))
2475 tag
= lookup_tag(oid
);
2477 if (!tag
|| parse_tag(tag
) || !tag
->tagged
)
2478 die("unable to pack objects reachable from tag %s",
2481 add_object_entry(&tag
->object
.oid
, OBJ_TAG
, NULL
, 0);
2483 if (tag
->tagged
->type
!= OBJ_TAG
)
2486 tag
= (struct tag
*)tag
->tagged
;
2490 static int add_ref_tag(const char *path
, const struct object_id
*oid
, int flag
, void *cb_data
)
2492 struct object_id peeled
;
2494 if (starts_with(path
, "refs/tags/") && /* is a tag? */
2495 !peel_ref(path
, &peeled
) && /* peelable? */
2496 packlist_find(&to_pack
, peeled
.hash
, NULL
)) /* object packed? */
2501 static void prepare_pack(int window
, int depth
)
2503 struct object_entry
**delta_list
;
2504 uint32_t i
, nr_deltas
;
2507 get_object_details();
2510 * If we're locally repacking then we need to be doubly careful
2511 * from now on in order to make sure no stealth corruption gets
2512 * propagated to the new pack. Clients receiving streamed packs
2513 * should validate everything they get anyway so no need to incur
2514 * the additional cost here in that case.
2516 if (!pack_to_stdout
)
2517 do_check_packed_object_crc
= 1;
2519 if (!to_pack
.nr_objects
|| !window
|| !depth
)
2522 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
2525 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2526 struct object_entry
*entry
= to_pack
.objects
+ i
;
2529 /* This happens if we decided to reuse existing
2530 * delta from a pack. "reuse_delta &&" is implied.
2534 if (!entry
->type_valid
||
2535 oe_size_less_than(&to_pack
, entry
, 50))
2538 if (entry
->no_try_delta
)
2541 if (!entry
->preferred_base
) {
2543 if (oe_type(entry
) < 0)
2544 die("unable to get type of object %s",
2545 oid_to_hex(&entry
->idx
.oid
));
2547 if (oe_type(entry
) < 0) {
2549 * This object is not found, but we
2550 * don't have to include it anyway.
2556 delta_list
[n
++] = entry
;
2559 if (nr_deltas
&& n
> 1) {
2560 unsigned nr_done
= 0;
2562 progress_state
= start_progress(_("Compressing objects"),
2564 QSORT(delta_list
, n
, type_size_sort
);
2565 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
2566 stop_progress(&progress_state
);
2567 if (nr_done
!= nr_deltas
)
2568 die("inconsistency with delta count");
2573 static int git_pack_config(const char *k
, const char *v
, void *cb
)
2575 if (!strcmp(k
, "pack.window")) {
2576 window
= git_config_int(k
, v
);
2579 if (!strcmp(k
, "pack.windowmemory")) {
2580 window_memory_limit
= git_config_ulong(k
, v
);
2583 if (!strcmp(k
, "pack.depth")) {
2584 depth
= git_config_int(k
, v
);
2587 if (!strcmp(k
, "pack.deltacachesize")) {
2588 max_delta_cache_size
= git_config_int(k
, v
);
2591 if (!strcmp(k
, "pack.deltacachelimit")) {
2592 cache_max_small_delta_size
= git_config_int(k
, v
);
2595 if (!strcmp(k
, "pack.writebitmaphashcache")) {
2596 if (git_config_bool(k
, v
))
2597 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
2599 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
2601 if (!strcmp(k
, "pack.usebitmaps")) {
2602 use_bitmap_index_default
= git_config_bool(k
, v
);
2605 if (!strcmp(k
, "pack.threads")) {
2606 delta_search_threads
= git_config_int(k
, v
);
2607 if (delta_search_threads
< 0)
2608 die("invalid number of threads specified (%d)",
2609 delta_search_threads
);
2611 if (delta_search_threads
!= 1) {
2612 warning("no threads support, ignoring %s", k
);
2613 delta_search_threads
= 0;
2618 if (!strcmp(k
, "pack.indexversion")) {
2619 pack_idx_opts
.version
= git_config_int(k
, v
);
2620 if (pack_idx_opts
.version
> 2)
2621 die("bad pack.indexversion=%"PRIu32
,
2622 pack_idx_opts
.version
);
2625 return git_default_config(k
, v
, cb
);
2628 static void read_object_list_from_stdin(void)
2630 char line
[GIT_MAX_HEXSZ
+ 1 + PATH_MAX
+ 2];
2631 struct object_id oid
;
2635 if (!fgets(line
, sizeof(line
), stdin
)) {
2639 die("fgets returned NULL, not EOF, not error!");
2645 if (line
[0] == '-') {
2646 if (get_oid_hex(line
+1, &oid
))
2647 die("expected edge object ID, got garbage:\n %s",
2649 add_preferred_base(&oid
);
2652 if (parse_oid_hex(line
, &oid
, &p
))
2653 die("expected object ID, got garbage:\n %s", line
);
2655 add_preferred_base_object(p
+ 1);
2656 add_object_entry(&oid
, OBJ_NONE
, p
+ 1, 0);
2660 /* Remember to update object flag allocation in object.h */
2661 #define OBJECT_ADDED (1u<<20)
2663 static void show_commit(struct commit
*commit
, void *data
)
2665 add_object_entry(&commit
->object
.oid
, OBJ_COMMIT
, NULL
, 0);
2666 commit
->object
.flags
|= OBJECT_ADDED
;
2668 if (write_bitmap_index
)
2669 index_commit_for_bitmap(commit
);
2672 static void show_object(struct object
*obj
, const char *name
, void *data
)
2674 add_preferred_base_object(name
);
2675 add_object_entry(&obj
->oid
, obj
->type
, name
, 0);
2676 obj
->flags
|= OBJECT_ADDED
;
2679 static void show_object__ma_allow_any(struct object
*obj
, const char *name
, void *data
)
2681 assert(arg_missing_action
== MA_ALLOW_ANY
);
2684 * Quietly ignore ALL missing objects. This avoids problems with
2685 * staging them now and getting an odd error later.
2687 if (!has_object_file(&obj
->oid
))
2690 show_object(obj
, name
, data
);
2693 static void show_object__ma_allow_promisor(struct object
*obj
, const char *name
, void *data
)
2695 assert(arg_missing_action
== MA_ALLOW_PROMISOR
);
2698 * Quietly ignore EXPECTED missing objects. This avoids problems with
2699 * staging them now and getting an odd error later.
2701 if (!has_object_file(&obj
->oid
) && is_promisor_object(&obj
->oid
))
2704 show_object(obj
, name
, data
);
2707 static int option_parse_missing_action(const struct option
*opt
,
2708 const char *arg
, int unset
)
2713 if (!strcmp(arg
, "error")) {
2714 arg_missing_action
= MA_ERROR
;
2715 fn_show_object
= show_object
;
2719 if (!strcmp(arg
, "allow-any")) {
2720 arg_missing_action
= MA_ALLOW_ANY
;
2721 fetch_if_missing
= 0;
2722 fn_show_object
= show_object__ma_allow_any
;
2726 if (!strcmp(arg
, "allow-promisor")) {
2727 arg_missing_action
= MA_ALLOW_PROMISOR
;
2728 fetch_if_missing
= 0;
2729 fn_show_object
= show_object__ma_allow_promisor
;
2733 die(_("invalid value for --missing"));
2737 static void show_edge(struct commit
*commit
)
2739 add_preferred_base(&commit
->object
.oid
);
2742 struct in_pack_object
{
2744 struct object
*object
;
2750 struct in_pack_object
*array
;
2753 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
2755 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->oid
.hash
, p
);
2756 in_pack
->array
[in_pack
->nr
].object
= object
;
2761 * Compare the objects in the offset order, in order to emulate the
2762 * "git rev-list --objects" output that produced the pack originally.
2764 static int ofscmp(const void *a_
, const void *b_
)
2766 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
2767 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
2769 if (a
->offset
< b
->offset
)
2771 else if (a
->offset
> b
->offset
)
2774 return oidcmp(&a
->object
->oid
, &b
->object
->oid
);
2777 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
2779 struct packed_git
*p
;
2780 struct in_pack in_pack
;
2783 memset(&in_pack
, 0, sizeof(in_pack
));
2785 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
2786 struct object_id oid
;
2789 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
2791 if (open_pack_index(p
))
2792 die("cannot open pack index");
2794 ALLOC_GROW(in_pack
.array
,
2795 in_pack
.nr
+ p
->num_objects
,
2798 for (i
= 0; i
< p
->num_objects
; i
++) {
2799 nth_packed_object_oid(&oid
, p
, i
);
2800 o
= lookup_unknown_object(oid
.hash
);
2801 if (!(o
->flags
& OBJECT_ADDED
))
2802 mark_in_pack_object(o
, p
, &in_pack
);
2803 o
->flags
|= OBJECT_ADDED
;
2808 QSORT(in_pack
.array
, in_pack
.nr
, ofscmp
);
2809 for (i
= 0; i
< in_pack
.nr
; i
++) {
2810 struct object
*o
= in_pack
.array
[i
].object
;
2811 add_object_entry(&o
->oid
, o
->type
, "", 0);
2814 free(in_pack
.array
);
2817 static int add_loose_object(const struct object_id
*oid
, const char *path
,
2820 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
2823 warning("loose object at %s could not be examined", path
);
2827 add_object_entry(oid
, type
, "", 0);
2832 * We actually don't even have to worry about reachability here.
2833 * add_object_entry will weed out duplicates, so we just add every
2834 * loose object we find.
2836 static void add_unreachable_loose_objects(void)
2838 for_each_loose_file_in_objdir(get_object_directory(),
2843 static int has_sha1_pack_kept_or_nonlocal(const struct object_id
*oid
)
2845 static struct packed_git
*last_found
= (void *)1;
2846 struct packed_git
*p
;
2848 p
= (last_found
!= (void *)1) ? last_found
:
2849 get_packed_git(the_repository
);
2852 if ((!p
->pack_local
|| p
->pack_keep
||
2853 p
->pack_keep_in_core
) &&
2854 find_pack_entry_one(oid
->hash
, p
)) {
2858 if (p
== last_found
)
2859 p
= get_packed_git(the_repository
);
2862 if (p
== last_found
)
2869 * Store a list of sha1s that are should not be discarded
2870 * because they are either written too recently, or are
2871 * reachable from another object that was.
2873 * This is filled by get_object_list.
2875 static struct oid_array recent_objects
;
2877 static int loosened_object_can_be_discarded(const struct object_id
*oid
,
2880 if (!unpack_unreachable_expiration
)
2882 if (mtime
> unpack_unreachable_expiration
)
2884 if (oid_array_lookup(&recent_objects
, oid
) >= 0)
2889 static void loosen_unused_packed_objects(struct rev_info
*revs
)
2891 struct packed_git
*p
;
2893 struct object_id oid
;
2895 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
2896 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
2899 if (open_pack_index(p
))
2900 die("cannot open pack index");
2902 for (i
= 0; i
< p
->num_objects
; i
++) {
2903 nth_packed_object_oid(&oid
, p
, i
);
2904 if (!packlist_find(&to_pack
, oid
.hash
, NULL
) &&
2905 !has_sha1_pack_kept_or_nonlocal(&oid
) &&
2906 !loosened_object_can_be_discarded(&oid
, p
->mtime
))
2907 if (force_object_loose(&oid
, p
->mtime
))
2908 die("unable to force loose object");
2914 * This tracks any options which pack-reuse code expects to be on, or which a
2915 * reader of the pack might not understand, and which would therefore prevent
2916 * blind reuse of what we have on disk.
2918 static int pack_options_allow_reuse(void)
2920 return pack_to_stdout
&&
2922 !ignore_packed_keep_on_disk
&&
2923 !ignore_packed_keep_in_core
&&
2924 (!local
|| !have_non_local_packs
) &&
2928 static int get_object_list_from_bitmap(struct rev_info
*revs
)
2930 if (prepare_bitmap_walk(revs
) < 0)
2933 if (pack_options_allow_reuse() &&
2934 !reuse_partial_packfile_from_bitmap(
2936 &reuse_packfile_objects
,
2937 &reuse_packfile_offset
)) {
2938 assert(reuse_packfile_objects
);
2939 nr_result
+= reuse_packfile_objects
;
2940 display_progress(progress_state
, nr_result
);
2943 traverse_bitmap_commit_list(&add_object_entry_from_bitmap
);
2947 static void record_recent_object(struct object
*obj
,
2951 oid_array_append(&recent_objects
, &obj
->oid
);
2954 static void record_recent_commit(struct commit
*commit
, void *data
)
2956 oid_array_append(&recent_objects
, &commit
->object
.oid
);
2959 static void get_object_list(int ac
, const char **av
)
2961 struct rev_info revs
;
2965 init_revisions(&revs
, NULL
);
2966 save_commit_buffer
= 0;
2967 setup_revisions(ac
, av
, &revs
, NULL
);
2969 /* make sure shallows are read */
2970 is_repository_shallow();
2972 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2973 int len
= strlen(line
);
2974 if (len
&& line
[len
- 1] == '\n')
2979 if (!strcmp(line
, "--not")) {
2980 flags
^= UNINTERESTING
;
2981 write_bitmap_index
= 0;
2984 if (starts_with(line
, "--shallow ")) {
2985 struct object_id oid
;
2986 if (get_oid_hex(line
+ 10, &oid
))
2987 die("not an SHA-1 '%s'", line
+ 10);
2988 register_shallow(&oid
);
2989 use_bitmap_index
= 0;
2992 die("not a rev '%s'", line
);
2994 if (handle_revision_arg(line
, &revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
2995 die("bad revision '%s'", line
);
2998 if (use_bitmap_index
&& !get_object_list_from_bitmap(&revs
))
3001 if (prepare_revision_walk(&revs
))
3002 die("revision walk setup failed");
3003 mark_edges_uninteresting(&revs
, show_edge
);
3005 if (!fn_show_object
)
3006 fn_show_object
= show_object
;
3007 traverse_commit_list_filtered(&filter_options
, &revs
,
3008 show_commit
, fn_show_object
, NULL
,
3011 if (unpack_unreachable_expiration
) {
3012 revs
.ignore_missing_links
= 1;
3013 if (add_unseen_recent_objects_to_traversal(&revs
,
3014 unpack_unreachable_expiration
))
3015 die("unable to add recent objects");
3016 if (prepare_revision_walk(&revs
))
3017 die("revision walk setup failed");
3018 traverse_commit_list(&revs
, record_recent_commit
,
3019 record_recent_object
, NULL
);
3022 if (keep_unreachable
)
3023 add_objects_in_unpacked_packs(&revs
);
3024 if (pack_loose_unreachable
)
3025 add_unreachable_loose_objects();
3026 if (unpack_unreachable
)
3027 loosen_unused_packed_objects(&revs
);
3029 oid_array_clear(&recent_objects
);
3032 static void add_extra_kept_packs(const struct string_list
*names
)
3034 struct packed_git
*p
;
3039 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
3040 const char *name
= basename(p
->pack_name
);
3046 for (i
= 0; i
< names
->nr
; i
++)
3047 if (!fspathcmp(name
, names
->items
[i
].string
))
3050 if (i
< names
->nr
) {
3051 p
->pack_keep_in_core
= 1;
3052 ignore_packed_keep_in_core
= 1;
3058 static int option_parse_index_version(const struct option
*opt
,
3059 const char *arg
, int unset
)
3062 const char *val
= arg
;
3063 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
3064 if (pack_idx_opts
.version
> 2)
3065 die(_("unsupported index version %s"), val
);
3066 if (*c
== ',' && c
[1])
3067 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
3068 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
3069 die(_("bad index version '%s'"), val
);
3073 static int option_parse_unpack_unreachable(const struct option
*opt
,
3074 const char *arg
, int unset
)
3077 unpack_unreachable
= 0;
3078 unpack_unreachable_expiration
= 0;
3081 unpack_unreachable
= 1;
3083 unpack_unreachable_expiration
= approxidate(arg
);
3088 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
3090 int use_internal_rev_list
= 0;
3093 int all_progress_implied
= 0;
3094 struct argv_array rp
= ARGV_ARRAY_INIT
;
3095 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
3096 int rev_list_index
= 0;
3097 struct string_list keep_pack_list
= STRING_LIST_INIT_NODUP
;
3098 struct option pack_objects_options
[] = {
3099 OPT_SET_INT('q', "quiet", &progress
,
3100 N_("do not show progress meter"), 0),
3101 OPT_SET_INT(0, "progress", &progress
,
3102 N_("show progress meter"), 1),
3103 OPT_SET_INT(0, "all-progress", &progress
,
3104 N_("show progress meter during object writing phase"), 2),
3105 OPT_BOOL(0, "all-progress-implied",
3106 &all_progress_implied
,
3107 N_("similar to --all-progress when progress meter is shown")),
3108 { OPTION_CALLBACK
, 0, "index-version", NULL
, N_("version[,offset]"),
3109 N_("write the pack index file in the specified idx format version"),
3110 0, option_parse_index_version
},
3111 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
3112 N_("maximum size of each output pack file")),
3113 OPT_BOOL(0, "local", &local
,
3114 N_("ignore borrowed objects from alternate object store")),
3115 OPT_BOOL(0, "incremental", &incremental
,
3116 N_("ignore packed objects")),
3117 OPT_INTEGER(0, "window", &window
,
3118 N_("limit pack window by objects")),
3119 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
3120 N_("limit pack window by memory in addition to object limit")),
3121 OPT_INTEGER(0, "depth", &depth
,
3122 N_("maximum length of delta chain allowed in the resulting pack")),
3123 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
3124 N_("reuse existing deltas")),
3125 OPT_BOOL(0, "reuse-object", &reuse_object
,
3126 N_("reuse existing objects")),
3127 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
3128 N_("use OFS_DELTA objects")),
3129 OPT_INTEGER(0, "threads", &delta_search_threads
,
3130 N_("use threads when searching for best delta matches")),
3131 OPT_BOOL(0, "non-empty", &non_empty
,
3132 N_("do not create an empty pack output")),
3133 OPT_BOOL(0, "revs", &use_internal_rev_list
,
3134 N_("read revision arguments from standard input")),
3135 { OPTION_SET_INT
, 0, "unpacked", &rev_list_unpacked
, NULL
,
3136 N_("limit the objects to those that are not yet packed"),
3137 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
3138 { OPTION_SET_INT
, 0, "all", &rev_list_all
, NULL
,
3139 N_("include objects reachable from any reference"),
3140 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
3141 { OPTION_SET_INT
, 0, "reflog", &rev_list_reflog
, NULL
,
3142 N_("include objects referred by reflog entries"),
3143 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
3144 { OPTION_SET_INT
, 0, "indexed-objects", &rev_list_index
, NULL
,
3145 N_("include objects referred to by the index"),
3146 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
3147 OPT_BOOL(0, "stdout", &pack_to_stdout
,
3148 N_("output pack to stdout")),
3149 OPT_BOOL(0, "include-tag", &include_tag
,
3150 N_("include tag objects that refer to objects to be packed")),
3151 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
3152 N_("keep unreachable objects")),
3153 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
3154 N_("pack loose unreachable objects")),
3155 { OPTION_CALLBACK
, 0, "unpack-unreachable", NULL
, N_("time"),
3156 N_("unpack unreachable objects newer than <time>"),
3157 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
},
3158 OPT_BOOL(0, "thin", &thin
,
3159 N_("create thin packs")),
3160 OPT_BOOL(0, "shallow", &shallow
,
3161 N_("create packs suitable for shallow fetches")),
3162 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk
,
3163 N_("ignore packs that have companion .keep file")),
3164 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list
, N_("name"),
3165 N_("ignore this pack")),
3166 OPT_INTEGER(0, "compression", &pack_compression_level
,
3167 N_("pack compression level")),
3168 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
3169 N_("do not hide commits by grafts"), 0),
3170 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
3171 N_("use a bitmap index if available to speed up counting objects")),
3172 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index
,
3173 N_("write a bitmap index together with the pack index")),
3174 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options
),
3175 { OPTION_CALLBACK
, 0, "missing", NULL
, N_("action"),
3176 N_("handling for missing objects"), PARSE_OPT_NONEG
,
3177 option_parse_missing_action
},
3178 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects
,
3179 N_("do not pack objects in promisor packfiles")),
3183 if (DFS_NUM_STATES
> (1 << OE_DFS_STATE_BITS
))
3184 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3186 check_replace_refs
= 0;
3188 reset_pack_idx_option(&pack_idx_opts
);
3189 git_config(git_pack_config
, NULL
);
3191 progress
= isatty(2);
3192 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
3196 base_name
= argv
[0];
3199 if (pack_to_stdout
!= !base_name
|| argc
)
3200 usage_with_options(pack_usage
, pack_objects_options
);
3202 if (depth
>= (1 << OE_DEPTH_BITS
)) {
3203 warning(_("delta chain depth %d is too deep, forcing %d"),
3204 depth
, (1 << OE_DEPTH_BITS
) - 1);
3205 depth
= (1 << OE_DEPTH_BITS
) - 1;
3207 if (cache_max_small_delta_size
>= (1U << OE_Z_DELTA_BITS
)) {
3208 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3209 (1U << OE_Z_DELTA_BITS
) - 1);
3210 cache_max_small_delta_size
= (1U << OE_Z_DELTA_BITS
) - 1;
3213 argv_array_push(&rp
, "pack-objects");
3215 use_internal_rev_list
= 1;
3216 argv_array_push(&rp
, shallow
3217 ? "--objects-edge-aggressive"
3218 : "--objects-edge");
3220 argv_array_push(&rp
, "--objects");
3223 use_internal_rev_list
= 1;
3224 argv_array_push(&rp
, "--all");
3226 if (rev_list_reflog
) {
3227 use_internal_rev_list
= 1;
3228 argv_array_push(&rp
, "--reflog");
3230 if (rev_list_index
) {
3231 use_internal_rev_list
= 1;
3232 argv_array_push(&rp
, "--indexed-objects");
3234 if (rev_list_unpacked
) {
3235 use_internal_rev_list
= 1;
3236 argv_array_push(&rp
, "--unpacked");
3239 if (exclude_promisor_objects
) {
3240 use_internal_rev_list
= 1;
3241 fetch_if_missing
= 0;
3242 argv_array_push(&rp
, "--exclude-promisor-objects");
3247 if (pack_compression_level
== -1)
3248 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
3249 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
3250 die("bad pack compression level %d", pack_compression_level
);
3252 if (!delta_search_threads
) /* --threads=0 means autodetect */
3253 delta_search_threads
= online_cpus();
3256 if (delta_search_threads
!= 1)
3257 warning("no threads support, ignoring --threads");
3259 if (!pack_to_stdout
&& !pack_size_limit
)
3260 pack_size_limit
= pack_size_limit_cfg
;
3261 if (pack_to_stdout
&& pack_size_limit
)
3262 die("--max-pack-size cannot be used to build a pack for transfer.");
3263 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
3264 warning("minimum pack size limit is 1 MiB");
3265 pack_size_limit
= 1024*1024;
3268 if (!pack_to_stdout
&& thin
)
3269 die("--thin cannot be used to build an indexable pack.");
3271 if (keep_unreachable
&& unpack_unreachable
)
3272 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3273 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
3274 unpack_unreachable_expiration
= 0;
3276 if (filter_options
.choice
) {
3277 if (!pack_to_stdout
)
3278 die("cannot use --filter without --stdout.");
3279 use_bitmap_index
= 0;
3283 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3285 * - to produce good pack (with bitmap index not-yet-packed objects are
3286 * packed in suboptimal order).
3288 * - to use more robust pack-generation codepath (avoiding possible
3289 * bugs in bitmap code and possible bitmap index corruption).
3291 if (!pack_to_stdout
)
3292 use_bitmap_index_default
= 0;
3294 if (use_bitmap_index
< 0)
3295 use_bitmap_index
= use_bitmap_index_default
;
3297 /* "hard" reasons not to use bitmaps; these just won't work at all */
3298 if (!use_internal_rev_list
|| (!pack_to_stdout
&& write_bitmap_index
) || is_repository_shallow())
3299 use_bitmap_index
= 0;
3301 if (pack_to_stdout
|| !rev_list_all
)
3302 write_bitmap_index
= 0;
3304 if (progress
&& all_progress_implied
)
3307 add_extra_kept_packs(&keep_pack_list
);
3308 if (ignore_packed_keep_on_disk
) {
3309 struct packed_git
*p
;
3310 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
)
3311 if (p
->pack_local
&& p
->pack_keep
)
3313 if (!p
) /* no keep-able packs found */
3314 ignore_packed_keep_on_disk
= 0;
3318 * unlike ignore_packed_keep_on_disk above, we do not
3319 * want to unset "local" based on looking at packs, as
3320 * it also covers non-local objects
3322 struct packed_git
*p
;
3323 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
3324 if (!p
->pack_local
) {
3325 have_non_local_packs
= 1;
3331 prepare_packing_data(&to_pack
);
3334 progress_state
= start_progress(_("Enumerating objects"), 0);
3335 if (!use_internal_rev_list
)
3336 read_object_list_from_stdin();
3338 get_object_list(rp
.argc
, rp
.argv
);
3339 argv_array_clear(&rp
);
3341 cleanup_preferred_base();
3342 if (include_tag
&& nr_result
)
3343 for_each_ref(add_ref_tag
, NULL
);
3344 stop_progress(&progress_state
);
3346 if (non_empty
&& !nr_result
)
3349 prepare_pack(window
, depth
);
3352 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
3353 " reused %"PRIu32
" (delta %"PRIu32
")\n",
3354 written
, written_delta
, reused
, reused_delta
);