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
;
282 const unsigned hashsz
= the_hash_algo
->rawsz
;
285 if (oe_type(entry
) == OBJ_BLOB
&&
286 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
) &&
287 (st
= open_istream(&entry
->idx
.oid
, &type
, &size
, NULL
)) != NULL
)
290 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
292 die(_("unable to read %s"),
293 oid_to_hex(&entry
->idx
.oid
));
296 * make sure no cached delta data remains from a
297 * previous attempt before a pack split occurred.
299 FREE_AND_NULL(entry
->delta_data
);
300 entry
->z_delta_size
= 0;
301 } else if (entry
->delta_data
) {
302 size
= DELTA_SIZE(entry
);
303 buf
= entry
->delta_data
;
304 entry
->delta_data
= NULL
;
305 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
306 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
308 buf
= get_delta(entry
);
309 size
= DELTA_SIZE(entry
);
310 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
311 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
314 if (st
) /* large blob case, just assume we don't compress well */
316 else if (entry
->z_delta_size
)
317 datalen
= entry
->z_delta_size
;
319 datalen
= do_compress(&buf
, size
);
322 * The object header is a byte of 'type' followed by zero or
323 * more bytes of length.
325 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
328 if (type
== OBJ_OFS_DELTA
) {
330 * Deltas with relative base contain an additional
331 * encoding of the relative offset for the delta
332 * base from this object's position in the pack.
334 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
335 unsigned pos
= sizeof(dheader
) - 1;
336 dheader
[pos
] = ofs
& 127;
338 dheader
[--pos
] = 128 | (--ofs
& 127);
339 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
345 hashwrite(f
, header
, hdrlen
);
346 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
347 hdrlen
+= sizeof(dheader
) - pos
;
348 } else if (type
== OBJ_REF_DELTA
) {
350 * Deltas with a base reference contain
351 * additional bytes for the base object ID.
353 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
359 hashwrite(f
, header
, hdrlen
);
360 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
363 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
369 hashwrite(f
, header
, hdrlen
);
372 datalen
= write_large_blob_data(st
, f
, &entry
->idx
.oid
);
375 hashwrite(f
, buf
, datalen
);
379 return hdrlen
+ datalen
;
382 /* Return 0 if we will bust the pack-size limit */
383 static off_t
write_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
384 unsigned long limit
, int usable_delta
)
386 struct packed_git
*p
= IN_PACK(entry
);
387 struct pack_window
*w_curs
= NULL
;
388 struct revindex_entry
*revidx
;
390 enum object_type type
= oe_type(entry
);
392 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
393 dheader
[MAX_PACK_OBJECT_HEADER
];
395 const unsigned hashsz
= the_hash_algo
->rawsz
;
396 unsigned long entry_size
= SIZE(entry
);
399 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
400 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
401 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
404 offset
= entry
->in_pack_offset
;
405 revidx
= find_pack_revindex(p
, offset
);
406 datalen
= revidx
[1].offset
- offset
;
407 if (!pack_to_stdout
&& p
->index_version
> 1 &&
408 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
)) {
409 error("bad packed object CRC for %s",
410 oid_to_hex(&entry
->idx
.oid
));
412 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
415 offset
+= entry
->in_pack_header_size
;
416 datalen
-= entry
->in_pack_header_size
;
418 if (!pack_to_stdout
&& p
->index_version
== 1 &&
419 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry_size
)) {
420 error("corrupt packed object for %s",
421 oid_to_hex(&entry
->idx
.oid
));
423 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
426 if (type
== OBJ_OFS_DELTA
) {
427 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
428 unsigned pos
= sizeof(dheader
) - 1;
429 dheader
[pos
] = ofs
& 127;
431 dheader
[--pos
] = 128 | (--ofs
& 127);
432 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
436 hashwrite(f
, header
, hdrlen
);
437 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
438 hdrlen
+= sizeof(dheader
) - pos
;
440 } else if (type
== OBJ_REF_DELTA
) {
441 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
445 hashwrite(f
, header
, hdrlen
);
446 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
450 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
454 hashwrite(f
, header
, hdrlen
);
456 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
459 return hdrlen
+ datalen
;
462 /* Return 0 if we will bust the pack-size limit */
463 static off_t
write_object(struct hashfile
*f
,
464 struct object_entry
*entry
,
469 int usable_delta
, to_reuse
;
474 /* apply size limit if limited packsize and not first object */
475 if (!pack_size_limit
|| !nr_written
)
477 else if (pack_size_limit
<= write_offset
)
479 * the earlier object did not fit the limit; avoid
480 * mistaking this with unlimited (i.e. limit = 0).
484 limit
= pack_size_limit
- write_offset
;
487 usable_delta
= 0; /* no delta */
488 else if (!pack_size_limit
)
489 usable_delta
= 1; /* unlimited packfile */
490 else if (DELTA(entry
)->idx
.offset
== (off_t
)-1)
491 usable_delta
= 0; /* base was written to another pack */
492 else if (DELTA(entry
)->idx
.offset
)
493 usable_delta
= 1; /* base already exists in this pack */
495 usable_delta
= 0; /* base could end up in another pack */
498 to_reuse
= 0; /* explicit */
499 else if (!IN_PACK(entry
))
500 to_reuse
= 0; /* can't reuse what we don't have */
501 else if (oe_type(entry
) == OBJ_REF_DELTA
||
502 oe_type(entry
) == OBJ_OFS_DELTA
)
503 /* check_object() decided it for us ... */
504 to_reuse
= usable_delta
;
505 /* ... but pack split may override that */
506 else if (oe_type(entry
) != entry
->in_pack_type
)
507 to_reuse
= 0; /* pack has delta which is unusable */
508 else if (DELTA(entry
))
509 to_reuse
= 0; /* we want to pack afresh */
511 to_reuse
= 1; /* we have it in-pack undeltified,
512 * and we do not need to deltify it.
516 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
518 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
526 entry
->idx
.crc32
= crc32_end(f
);
530 enum write_one_status
{
531 WRITE_ONE_SKIP
= -1, /* already written */
532 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
533 WRITE_ONE_WRITTEN
= 1, /* normal */
534 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
537 static enum write_one_status
write_one(struct hashfile
*f
,
538 struct object_entry
*e
,
545 * we set offset to 1 (which is an impossible value) to mark
546 * the fact that this object is involved in "write its base
547 * first before writing a deltified object" recursion.
549 recursing
= (e
->idx
.offset
== 1);
551 warning("recursive delta detected for object %s",
552 oid_to_hex(&e
->idx
.oid
));
553 return WRITE_ONE_RECURSIVE
;
554 } else if (e
->idx
.offset
|| e
->preferred_base
) {
555 /* offset is non zero if object is written already. */
556 return WRITE_ONE_SKIP
;
559 /* if we are deltified, write out base object first. */
561 e
->idx
.offset
= 1; /* now recurse */
562 switch (write_one(f
, DELTA(e
), offset
)) {
563 case WRITE_ONE_RECURSIVE
:
564 /* we cannot depend on this one */
569 case WRITE_ONE_BREAK
:
570 e
->idx
.offset
= recursing
;
571 return WRITE_ONE_BREAK
;
575 e
->idx
.offset
= *offset
;
576 size
= write_object(f
, e
, *offset
);
578 e
->idx
.offset
= recursing
;
579 return WRITE_ONE_BREAK
;
581 written_list
[nr_written
++] = &e
->idx
;
583 /* make sure off_t is sufficiently large not to wrap */
584 if (signed_add_overflows(*offset
, size
))
585 die("pack too large for current definition of off_t");
587 return WRITE_ONE_WRITTEN
;
590 static int mark_tagged(const char *path
, const struct object_id
*oid
, int flag
,
593 struct object_id peeled
;
594 struct object_entry
*entry
= packlist_find(&to_pack
, oid
->hash
, NULL
);
598 if (!peel_ref(path
, &peeled
)) {
599 entry
= packlist_find(&to_pack
, peeled
.hash
, NULL
);
606 static inline void add_to_write_order(struct object_entry
**wo
,
608 struct object_entry
*e
)
616 static void add_descendants_to_write_order(struct object_entry
**wo
,
618 struct object_entry
*e
)
620 int add_to_order
= 1;
623 struct object_entry
*s
;
624 /* add this node... */
625 add_to_write_order(wo
, endp
, e
);
626 /* all its siblings... */
627 for (s
= DELTA_SIBLING(e
); s
; s
= DELTA_SIBLING(s
)) {
628 add_to_write_order(wo
, endp
, s
);
631 /* drop down a level to add left subtree nodes if possible */
632 if (DELTA_CHILD(e
)) {
637 /* our sibling might have some children, it is next */
638 if (DELTA_SIBLING(e
)) {
639 e
= DELTA_SIBLING(e
);
642 /* go back to our parent node */
644 while (e
&& !DELTA_SIBLING(e
)) {
645 /* we're on the right side of a subtree, keep
646 * going up until we can go right again */
650 /* done- we hit our original root node */
653 /* pass it off to sibling at this level */
654 e
= DELTA_SIBLING(e
);
659 static void add_family_to_write_order(struct object_entry
**wo
,
661 struct object_entry
*e
)
663 struct object_entry
*root
;
665 for (root
= e
; DELTA(root
); root
= DELTA(root
))
667 add_descendants_to_write_order(wo
, endp
, root
);
670 static void compute_layer_order(struct object_entry
**wo
, unsigned int *wo_end
)
672 unsigned int i
, last_untagged
;
673 struct object_entry
*objects
= to_pack
.objects
;
675 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
676 if (objects
[i
].tagged
)
678 add_to_write_order(wo
, wo_end
, &objects
[i
]);
683 * Then fill all the tagged tips.
685 for (; i
< to_pack
.nr_objects
; i
++) {
686 if (objects
[i
].tagged
)
687 add_to_write_order(wo
, wo_end
, &objects
[i
]);
691 * And then all remaining commits and tags.
693 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
694 if (oe_type(&objects
[i
]) != OBJ_COMMIT
&&
695 oe_type(&objects
[i
]) != OBJ_TAG
)
697 add_to_write_order(wo
, wo_end
, &objects
[i
]);
701 * And then all the trees.
703 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
704 if (oe_type(&objects
[i
]) != OBJ_TREE
)
706 add_to_write_order(wo
, wo_end
, &objects
[i
]);
710 * Finally all the rest in really tight order
712 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
713 if (!objects
[i
].filled
)
714 add_family_to_write_order(wo
, wo_end
, &objects
[i
]);
718 static struct object_entry
**compute_write_order(void)
720 unsigned int i
, wo_end
;
722 struct object_entry
**wo
;
723 struct object_entry
*objects
= to_pack
.objects
;
725 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
726 objects
[i
].tagged
= 0;
727 objects
[i
].filled
= 0;
728 SET_DELTA_CHILD(&objects
[i
], NULL
);
729 SET_DELTA_SIBLING(&objects
[i
], NULL
);
733 * Fully connect delta_child/delta_sibling network.
734 * Make sure delta_sibling is sorted in the original
737 for (i
= to_pack
.nr_objects
; i
> 0;) {
738 struct object_entry
*e
= &objects
[--i
];
741 /* Mark me as the first child */
742 e
->delta_sibling_idx
= DELTA(e
)->delta_child_idx
;
743 SET_DELTA_CHILD(DELTA(e
), e
);
747 * Mark objects that are at the tip of tags.
749 for_each_tag_ref(mark_tagged
, NULL
);
752 * Give the objects in the original recency order until
753 * we see a tagged tip.
755 ALLOC_ARRAY(wo
, to_pack
.nr_objects
);
758 compute_layer_order(wo
, &wo_end
);
760 if (wo_end
!= to_pack
.nr_objects
)
761 die("ordered %u objects, expected %"PRIu32
, wo_end
, to_pack
.nr_objects
);
766 static off_t
write_reused_pack(struct hashfile
*f
)
768 unsigned char buffer
[8192];
769 off_t to_write
, total
;
772 if (!is_pack_valid(reuse_packfile
))
773 die("packfile is invalid: %s", reuse_packfile
->pack_name
);
775 fd
= git_open(reuse_packfile
->pack_name
);
777 die_errno("unable to open packfile for reuse: %s",
778 reuse_packfile
->pack_name
);
780 if (lseek(fd
, sizeof(struct pack_header
), SEEK_SET
) == -1)
781 die_errno("unable to seek in reused packfile");
783 if (reuse_packfile_offset
< 0)
784 reuse_packfile_offset
= reuse_packfile
->pack_size
- the_hash_algo
->rawsz
;
786 total
= to_write
= reuse_packfile_offset
- sizeof(struct pack_header
);
789 int read_pack
= xread(fd
, buffer
, sizeof(buffer
));
792 die_errno("unable to read from reused packfile");
794 if (read_pack
> to_write
)
795 read_pack
= to_write
;
797 hashwrite(f
, buffer
, read_pack
);
798 to_write
-= read_pack
;
801 * We don't know the actual number of objects written,
802 * only how many bytes written, how many bytes total, and
803 * how many objects total. So we can fake it by pretending all
804 * objects we are writing are the same size. This gives us a
805 * smooth progress meter, and at the end it matches the true
808 written
= reuse_packfile_objects
*
809 (((double)(total
- to_write
)) / total
);
810 display_progress(progress_state
, written
);
814 written
= reuse_packfile_objects
;
815 display_progress(progress_state
, written
);
816 return reuse_packfile_offset
- sizeof(struct pack_header
);
819 static const char no_split_warning
[] = N_(
820 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
823 static void write_pack_file(void)
828 uint32_t nr_remaining
= nr_result
;
829 time_t last_mtime
= 0;
830 struct object_entry
**write_order
;
832 if (progress
> pack_to_stdout
)
833 progress_state
= start_progress(_("Writing objects"), nr_result
);
834 ALLOC_ARRAY(written_list
, to_pack
.nr_objects
);
835 write_order
= compute_write_order();
838 struct object_id oid
;
839 char *pack_tmp_name
= NULL
;
842 f
= hashfd_throughput(1, "<stdout>", progress_state
);
844 f
= create_tmp_packfile(&pack_tmp_name
);
846 offset
= write_pack_header(f
, nr_remaining
);
848 if (reuse_packfile
) {
850 assert(pack_to_stdout
);
852 packfile_size
= write_reused_pack(f
);
853 offset
+= packfile_size
;
857 for (; i
< to_pack
.nr_objects
; i
++) {
858 struct object_entry
*e
= write_order
[i
];
859 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
861 display_progress(progress_state
, written
);
865 * Did we write the wrong # entries in the header?
866 * If so, rewrite it like in fast-import
868 if (pack_to_stdout
) {
869 finalize_hashfile(f
, oid
.hash
, CSUM_HASH_IN_STREAM
| CSUM_CLOSE
);
870 } else if (nr_written
== nr_remaining
) {
871 finalize_hashfile(f
, oid
.hash
, CSUM_HASH_IN_STREAM
| CSUM_FSYNC
| CSUM_CLOSE
);
873 int fd
= finalize_hashfile(f
, oid
.hash
, 0);
874 fixup_pack_header_footer(fd
, oid
.hash
, pack_tmp_name
,
875 nr_written
, oid
.hash
, offset
);
877 if (write_bitmap_index
) {
878 warning(_(no_split_warning
));
879 write_bitmap_index
= 0;
883 if (!pack_to_stdout
) {
885 struct strbuf tmpname
= STRBUF_INIT
;
888 * Packs are runtime accessed in their mtime
889 * order since newer packs are more likely to contain
890 * younger objects. So if we are creating multiple
891 * packs then we should modify the mtime of later ones
892 * to preserve this property.
894 if (stat(pack_tmp_name
, &st
) < 0) {
895 warning_errno("failed to stat %s", pack_tmp_name
);
896 } else if (!last_mtime
) {
897 last_mtime
= st
.st_mtime
;
900 utb
.actime
= st
.st_atime
;
901 utb
.modtime
= --last_mtime
;
902 if (utime(pack_tmp_name
, &utb
) < 0)
903 warning_errno("failed utime() on %s", pack_tmp_name
);
906 strbuf_addf(&tmpname
, "%s-", base_name
);
908 if (write_bitmap_index
) {
909 bitmap_writer_set_checksum(oid
.hash
);
910 bitmap_writer_build_type_index(
911 &to_pack
, written_list
, nr_written
);
914 finish_tmp_packfile(&tmpname
, pack_tmp_name
,
915 written_list
, nr_written
,
916 &pack_idx_opts
, oid
.hash
);
918 if (write_bitmap_index
) {
919 strbuf_addf(&tmpname
, "%s.bitmap", oid_to_hex(&oid
));
921 stop_progress(&progress_state
);
923 bitmap_writer_show_progress(progress
);
924 bitmap_writer_reuse_bitmaps(&to_pack
);
925 bitmap_writer_select_commits(indexed_commits
, indexed_commits_nr
, -1);
926 bitmap_writer_build(&to_pack
);
927 bitmap_writer_finish(written_list
, nr_written
,
928 tmpname
.buf
, write_bitmap_options
);
929 write_bitmap_index
= 0;
932 strbuf_release(&tmpname
);
934 puts(oid_to_hex(&oid
));
937 /* mark written objects as written to previous pack */
938 for (j
= 0; j
< nr_written
; j
++) {
939 written_list
[j
]->offset
= (off_t
)-1;
941 nr_remaining
-= nr_written
;
942 } while (nr_remaining
&& i
< to_pack
.nr_objects
);
946 stop_progress(&progress_state
);
947 if (written
!= nr_result
)
948 die("wrote %"PRIu32
" objects while expecting %"PRIu32
,
952 static int no_try_delta(const char *path
)
954 static struct attr_check
*check
;
957 check
= attr_check_initl("delta", NULL
);
958 if (git_check_attr(path
, check
))
960 if (ATTR_FALSE(check
->items
[0].value
))
966 * When adding an object, check whether we have already added it
967 * to our packing list. If so, we can skip. However, if we are
968 * being asked to excludei t, but the previous mention was to include
969 * it, make sure to adjust its flags and tweak our numbers accordingly.
971 * As an optimization, we pass out the index position where we would have
972 * found the item, since that saves us from having to look it up again a
973 * few lines later when we want to add the new entry.
975 static int have_duplicate_entry(const struct object_id
*oid
,
979 struct object_entry
*entry
;
981 entry
= packlist_find(&to_pack
, oid
->hash
, index_pos
);
986 if (!entry
->preferred_base
)
988 entry
->preferred_base
= 1;
994 static int want_found_object(int exclude
, struct packed_git
*p
)
1002 * When asked to do --local (do not include an object that appears in a
1003 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1004 * an object that appears in a pack marked with .keep), finding a pack
1005 * that matches the criteria is sufficient for us to decide to omit it.
1006 * However, even if this pack does not satisfy the criteria, we need to
1007 * make sure no copy of this object appears in _any_ pack that makes us
1008 * to omit the object, so we need to check all the packs.
1010 * We can however first check whether these options can possible matter;
1011 * if they do not matter we know we want the object in generated pack.
1012 * Otherwise, we signal "-1" at the end to tell the caller that we do
1013 * not know either way, and it needs to check more packs.
1015 if (!ignore_packed_keep_on_disk
&&
1016 !ignore_packed_keep_in_core
&&
1017 (!local
|| !have_non_local_packs
))
1020 if (local
&& !p
->pack_local
)
1022 if (p
->pack_local
&&
1023 ((ignore_packed_keep_on_disk
&& p
->pack_keep
) ||
1024 (ignore_packed_keep_in_core
&& p
->pack_keep_in_core
)))
1027 /* we don't know yet; keep looking for more packs */
1032 * Check whether we want the object in the pack (e.g., we do not want
1033 * objects found in non-local stores if the "--local" option was used).
1035 * If the caller already knows an existing pack it wants to take the object
1036 * from, that is passed in *found_pack and *found_offset; otherwise this
1037 * function finds if there is any pack that has the object and returns the pack
1038 * and its offset in these variables.
1040 static int want_object_in_pack(const struct object_id
*oid
,
1042 struct packed_git
**found_pack
,
1043 off_t
*found_offset
)
1046 struct list_head
*pos
;
1048 if (!exclude
&& local
&& has_loose_object_nonlocal(oid
))
1052 * If we already know the pack object lives in, start checks from that
1053 * pack - in the usual case when neither --local was given nor .keep files
1054 * are present we will determine the answer right now.
1057 want
= want_found_object(exclude
, *found_pack
);
1061 list_for_each(pos
, get_packed_git_mru(the_repository
)) {
1062 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1065 if (p
== *found_pack
)
1066 offset
= *found_offset
;
1068 offset
= find_pack_entry_one(oid
->hash
, p
);
1072 if (!is_pack_valid(p
))
1074 *found_offset
= offset
;
1077 want
= want_found_object(exclude
, p
);
1078 if (!exclude
&& want
> 0)
1080 get_packed_git_mru(the_repository
));
1089 static void create_object_entry(const struct object_id
*oid
,
1090 enum object_type type
,
1095 struct packed_git
*found_pack
,
1098 struct object_entry
*entry
;
1100 entry
= packlist_alloc(&to_pack
, oid
->hash
, index_pos
);
1102 oe_set_type(entry
, type
);
1104 entry
->preferred_base
= 1;
1108 oe_set_in_pack(&to_pack
, entry
, found_pack
);
1109 entry
->in_pack_offset
= found_offset
;
1112 entry
->no_try_delta
= no_try_delta
;
1115 static const char no_closure_warning
[] = N_(
1116 "disabling bitmap writing, as some objects are not being packed"
1119 static int add_object_entry(const struct object_id
*oid
, enum object_type type
,
1120 const char *name
, int exclude
)
1122 struct packed_git
*found_pack
= NULL
;
1123 off_t found_offset
= 0;
1126 display_progress(progress_state
, ++nr_seen
);
1128 if (have_duplicate_entry(oid
, exclude
, &index_pos
))
1131 if (!want_object_in_pack(oid
, exclude
, &found_pack
, &found_offset
)) {
1132 /* The pack is missing an object, so it will not have closure */
1133 if (write_bitmap_index
) {
1134 warning(_(no_closure_warning
));
1135 write_bitmap_index
= 0;
1140 create_object_entry(oid
, type
, pack_name_hash(name
),
1141 exclude
, name
&& no_try_delta(name
),
1142 index_pos
, found_pack
, found_offset
);
1146 static int add_object_entry_from_bitmap(const struct object_id
*oid
,
1147 enum object_type type
,
1148 int flags
, uint32_t name_hash
,
1149 struct packed_git
*pack
, off_t offset
)
1153 display_progress(progress_state
, ++nr_seen
);
1155 if (have_duplicate_entry(oid
, 0, &index_pos
))
1158 if (!want_object_in_pack(oid
, 0, &pack
, &offset
))
1161 create_object_entry(oid
, type
, name_hash
, 0, 0, index_pos
, pack
, offset
);
1165 struct pbase_tree_cache
{
1166 struct object_id oid
;
1170 unsigned long tree_size
;
1173 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1174 static int pbase_tree_cache_ix(const struct object_id
*oid
)
1176 return oid
->hash
[0] % ARRAY_SIZE(pbase_tree_cache
);
1178 static int pbase_tree_cache_ix_incr(int ix
)
1180 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1183 static struct pbase_tree
{
1184 struct pbase_tree
*next
;
1185 /* This is a phony "cache" entry; we are not
1186 * going to evict it or find it through _get()
1187 * mechanism -- this is for the toplevel node that
1188 * would almost always change with any commit.
1190 struct pbase_tree_cache pcache
;
1193 static struct pbase_tree_cache
*pbase_tree_get(const struct object_id
*oid
)
1195 struct pbase_tree_cache
*ent
, *nent
;
1198 enum object_type type
;
1200 int my_ix
= pbase_tree_cache_ix(oid
);
1201 int available_ix
= -1;
1203 /* pbase-tree-cache acts as a limited hashtable.
1204 * your object will be found at your index or within a few
1205 * slots after that slot if it is cached.
1207 for (neigh
= 0; neigh
< 8; neigh
++) {
1208 ent
= pbase_tree_cache
[my_ix
];
1209 if (ent
&& !oidcmp(&ent
->oid
, oid
)) {
1213 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1214 ((0 <= available_ix
) &&
1215 (!ent
&& pbase_tree_cache
[available_ix
])))
1216 available_ix
= my_ix
;
1219 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1222 /* Did not find one. Either we got a bogus request or
1223 * we need to read and perhaps cache.
1225 data
= read_object_file(oid
, &type
, &size
);
1228 if (type
!= OBJ_TREE
) {
1233 /* We need to either cache or return a throwaway copy */
1235 if (available_ix
< 0)
1238 ent
= pbase_tree_cache
[available_ix
];
1239 my_ix
= available_ix
;
1243 nent
= xmalloc(sizeof(*nent
));
1244 nent
->temporary
= (available_ix
< 0);
1247 /* evict and reuse */
1248 free(ent
->tree_data
);
1251 oidcpy(&nent
->oid
, oid
);
1252 nent
->tree_data
= data
;
1253 nent
->tree_size
= size
;
1255 if (!nent
->temporary
)
1256 pbase_tree_cache
[my_ix
] = nent
;
1260 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1262 if (!cache
->temporary
) {
1266 free(cache
->tree_data
);
1270 static int name_cmp_len(const char *name
)
1273 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
1278 static void add_pbase_object(struct tree_desc
*tree
,
1281 const char *fullname
)
1283 struct name_entry entry
;
1286 while (tree_entry(tree
,&entry
)) {
1287 if (S_ISGITLINK(entry
.mode
))
1289 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1290 memcmp(name
, entry
.path
, cmplen
);
1295 if (name
[cmplen
] != '/') {
1296 add_object_entry(entry
.oid
,
1297 object_type(entry
.mode
),
1301 if (S_ISDIR(entry
.mode
)) {
1302 struct tree_desc sub
;
1303 struct pbase_tree_cache
*tree
;
1304 const char *down
= name
+cmplen
+1;
1305 int downlen
= name_cmp_len(down
);
1307 tree
= pbase_tree_get(entry
.oid
);
1310 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1312 add_pbase_object(&sub
, down
, downlen
, fullname
);
1313 pbase_tree_put(tree
);
1318 static unsigned *done_pbase_paths
;
1319 static int done_pbase_paths_num
;
1320 static int done_pbase_paths_alloc
;
1321 static int done_pbase_path_pos(unsigned hash
)
1324 int hi
= done_pbase_paths_num
;
1326 int mi
= lo
+ (hi
- lo
) / 2;
1327 if (done_pbase_paths
[mi
] == hash
)
1329 if (done_pbase_paths
[mi
] < hash
)
1337 static int check_pbase_path(unsigned hash
)
1339 int pos
= done_pbase_path_pos(hash
);
1343 ALLOC_GROW(done_pbase_paths
,
1344 done_pbase_paths_num
+ 1,
1345 done_pbase_paths_alloc
);
1346 done_pbase_paths_num
++;
1347 if (pos
< done_pbase_paths_num
)
1348 MOVE_ARRAY(done_pbase_paths
+ pos
+ 1, done_pbase_paths
+ pos
,
1349 done_pbase_paths_num
- pos
- 1);
1350 done_pbase_paths
[pos
] = hash
;
1354 static void add_preferred_base_object(const char *name
)
1356 struct pbase_tree
*it
;
1358 unsigned hash
= pack_name_hash(name
);
1360 if (!num_preferred_base
|| check_pbase_path(hash
))
1363 cmplen
= name_cmp_len(name
);
1364 for (it
= pbase_tree
; it
; it
= it
->next
) {
1366 add_object_entry(&it
->pcache
.oid
, OBJ_TREE
, NULL
, 1);
1369 struct tree_desc tree
;
1370 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1371 add_pbase_object(&tree
, name
, cmplen
, name
);
1376 static void add_preferred_base(struct object_id
*oid
)
1378 struct pbase_tree
*it
;
1381 struct object_id tree_oid
;
1383 if (window
<= num_preferred_base
++)
1386 data
= read_object_with_reference(oid
, tree_type
, &size
, &tree_oid
);
1390 for (it
= pbase_tree
; it
; it
= it
->next
) {
1391 if (!oidcmp(&it
->pcache
.oid
, &tree_oid
)) {
1397 it
= xcalloc(1, sizeof(*it
));
1398 it
->next
= pbase_tree
;
1401 oidcpy(&it
->pcache
.oid
, &tree_oid
);
1402 it
->pcache
.tree_data
= data
;
1403 it
->pcache
.tree_size
= size
;
1406 static void cleanup_preferred_base(void)
1408 struct pbase_tree
*it
;
1414 struct pbase_tree
*tmp
= it
;
1416 free(tmp
->pcache
.tree_data
);
1420 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1421 if (!pbase_tree_cache
[i
])
1423 free(pbase_tree_cache
[i
]->tree_data
);
1424 FREE_AND_NULL(pbase_tree_cache
[i
]);
1427 FREE_AND_NULL(done_pbase_paths
);
1428 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1431 static void check_object(struct object_entry
*entry
)
1433 unsigned long canonical_size
;
1435 if (IN_PACK(entry
)) {
1436 struct packed_git
*p
= IN_PACK(entry
);
1437 struct pack_window
*w_curs
= NULL
;
1438 const unsigned char *base_ref
= NULL
;
1439 struct object_entry
*base_entry
;
1440 unsigned long used
, used_0
;
1441 unsigned long avail
;
1443 unsigned char *buf
, c
;
1444 enum object_type type
;
1445 unsigned long in_pack_size
;
1447 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1450 * We want in_pack_type even if we do not reuse delta
1451 * since non-delta representations could still be reused.
1453 used
= unpack_object_header_buffer(buf
, avail
,
1460 BUG("invalid type %d", type
);
1461 entry
->in_pack_type
= type
;
1464 * Determine if this is a delta and if so whether we can
1465 * reuse it or not. Otherwise let's find out as cheaply as
1466 * possible what the actual type and size for this object is.
1468 switch (entry
->in_pack_type
) {
1470 /* Not a delta hence we've already got all we need. */
1471 oe_set_type(entry
, entry
->in_pack_type
);
1472 SET_SIZE(entry
, in_pack_size
);
1473 entry
->in_pack_header_size
= used
;
1474 if (oe_type(entry
) < OBJ_COMMIT
|| oe_type(entry
) > OBJ_BLOB
)
1476 unuse_pack(&w_curs
);
1479 if (reuse_delta
&& !entry
->preferred_base
)
1480 base_ref
= use_pack(p
, &w_curs
,
1481 entry
->in_pack_offset
+ used
, NULL
);
1482 entry
->in_pack_header_size
= used
+ the_hash_algo
->rawsz
;
1485 buf
= use_pack(p
, &w_curs
,
1486 entry
->in_pack_offset
+ used
, NULL
);
1492 if (!ofs
|| MSB(ofs
, 7)) {
1493 error("delta base offset overflow in pack for %s",
1494 oid_to_hex(&entry
->idx
.oid
));
1498 ofs
= (ofs
<< 7) + (c
& 127);
1500 ofs
= entry
->in_pack_offset
- ofs
;
1501 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1502 error("delta base offset out of bound for %s",
1503 oid_to_hex(&entry
->idx
.oid
));
1506 if (reuse_delta
&& !entry
->preferred_base
) {
1507 struct revindex_entry
*revidx
;
1508 revidx
= find_pack_revindex(p
, ofs
);
1511 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1513 entry
->in_pack_header_size
= used
+ used_0
;
1517 if (base_ref
&& (base_entry
= packlist_find(&to_pack
, base_ref
, NULL
))) {
1519 * If base_ref was set above that means we wish to
1520 * reuse delta data, and we even found that base
1521 * in the list of objects we want to pack. Goodie!
1523 * Depth value does not matter - find_deltas() will
1524 * never consider reused delta as the base object to
1525 * deltify other objects against, in order to avoid
1528 oe_set_type(entry
, entry
->in_pack_type
);
1529 SET_SIZE(entry
, in_pack_size
); /* delta size */
1530 SET_DELTA(entry
, base_entry
);
1531 SET_DELTA_SIZE(entry
, in_pack_size
);
1532 entry
->delta_sibling_idx
= base_entry
->delta_child_idx
;
1533 SET_DELTA_CHILD(base_entry
, entry
);
1534 unuse_pack(&w_curs
);
1538 if (oe_type(entry
)) {
1542 * This must be a delta and we already know what the
1543 * final object type is. Let's extract the actual
1544 * object size from the delta header.
1546 delta_pos
= entry
->in_pack_offset
+ entry
->in_pack_header_size
;
1547 canonical_size
= get_size_from_delta(p
, &w_curs
, delta_pos
);
1548 if (canonical_size
== 0)
1550 SET_SIZE(entry
, canonical_size
);
1551 unuse_pack(&w_curs
);
1556 * No choice but to fall back to the recursive delta walk
1557 * with sha1_object_info() to find about the object type
1561 unuse_pack(&w_curs
);
1565 oid_object_info(the_repository
, &entry
->idx
.oid
, &canonical_size
));
1566 if (entry
->type_valid
) {
1567 SET_SIZE(entry
, canonical_size
);
1570 * Bad object type is checked in prepare_pack(). This is
1571 * to permit a missing preferred base object to be ignored
1572 * as a preferred base. Doing so can result in a larger
1573 * pack file, but the transfer will still take place.
1578 static int pack_offset_sort(const void *_a
, const void *_b
)
1580 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1581 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1582 const struct packed_git
*a_in_pack
= IN_PACK(a
);
1583 const struct packed_git
*b_in_pack
= IN_PACK(b
);
1585 /* avoid filesystem trashing with loose objects */
1586 if (!a_in_pack
&& !b_in_pack
)
1587 return oidcmp(&a
->idx
.oid
, &b
->idx
.oid
);
1589 if (a_in_pack
< b_in_pack
)
1591 if (a_in_pack
> b_in_pack
)
1593 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1594 (a
->in_pack_offset
> b
->in_pack_offset
);
1598 * Drop an on-disk delta we were planning to reuse. Naively, this would
1599 * just involve blanking out the "delta" field, but we have to deal
1600 * with some extra book-keeping:
1602 * 1. Removing ourselves from the delta_sibling linked list.
1604 * 2. Updating our size/type to the non-delta representation. These were
1605 * either not recorded initially (size) or overwritten with the delta type
1606 * (type) when check_object() decided to reuse the delta.
1608 * 3. Resetting our delta depth, as we are now a base object.
1610 static void drop_reused_delta(struct object_entry
*entry
)
1612 unsigned *idx
= &to_pack
.objects
[entry
->delta_idx
- 1].delta_child_idx
;
1613 struct object_info oi
= OBJECT_INFO_INIT
;
1614 enum object_type type
;
1618 struct object_entry
*oe
= &to_pack
.objects
[*idx
- 1];
1621 *idx
= oe
->delta_sibling_idx
;
1623 idx
= &oe
->delta_sibling_idx
;
1625 SET_DELTA(entry
, NULL
);
1630 if (packed_object_info(the_repository
, IN_PACK(entry
), entry
->in_pack_offset
, &oi
) < 0) {
1632 * We failed to get the info from this pack for some reason;
1633 * fall back to sha1_object_info, which may find another copy.
1634 * And if that fails, the error will be recorded in oe_type(entry)
1635 * and dealt with in prepare_pack().
1638 oid_object_info(the_repository
, &entry
->idx
.oid
, &size
));
1640 oe_set_type(entry
, type
);
1642 SET_SIZE(entry
, size
);
1646 * Follow the chain of deltas from this entry onward, throwing away any links
1647 * that cause us to hit a cycle (as determined by the DFS state flags in
1650 * We also detect too-long reused chains that would violate our --depth
1653 static void break_delta_chains(struct object_entry
*entry
)
1656 * The actual depth of each object we will write is stored as an int,
1657 * as it cannot exceed our int "depth" limit. But before we break
1658 * changes based no that limit, we may potentially go as deep as the
1659 * number of objects, which is elsewhere bounded to a uint32_t.
1661 uint32_t total_depth
;
1662 struct object_entry
*cur
, *next
;
1664 for (cur
= entry
, total_depth
= 0;
1666 cur
= DELTA(cur
), total_depth
++) {
1667 if (cur
->dfs_state
== DFS_DONE
) {
1669 * We've already seen this object and know it isn't
1670 * part of a cycle. We do need to append its depth
1673 total_depth
+= cur
->depth
;
1678 * We break cycles before looping, so an ACTIVE state (or any
1679 * other cruft which made its way into the state variable)
1682 if (cur
->dfs_state
!= DFS_NONE
)
1683 BUG("confusing delta dfs state in first pass: %d",
1687 * Now we know this is the first time we've seen the object. If
1688 * it's not a delta, we're done traversing, but we'll mark it
1689 * done to save time on future traversals.
1692 cur
->dfs_state
= DFS_DONE
;
1697 * Mark ourselves as active and see if the next step causes
1698 * us to cycle to another active object. It's important to do
1699 * this _before_ we loop, because it impacts where we make the
1700 * cut, and thus how our total_depth counter works.
1701 * E.g., We may see a partial loop like:
1703 * A -> B -> C -> D -> B
1705 * Cutting B->C breaks the cycle. But now the depth of A is
1706 * only 1, and our total_depth counter is at 3. The size of the
1707 * error is always one less than the size of the cycle we
1708 * broke. Commits C and D were "lost" from A's chain.
1710 * If we instead cut D->B, then the depth of A is correct at 3.
1711 * We keep all commits in the chain that we examined.
1713 cur
->dfs_state
= DFS_ACTIVE
;
1714 if (DELTA(cur
)->dfs_state
== DFS_ACTIVE
) {
1715 drop_reused_delta(cur
);
1716 cur
->dfs_state
= DFS_DONE
;
1722 * And now that we've gone all the way to the bottom of the chain, we
1723 * need to clear the active flags and set the depth fields as
1724 * appropriate. Unlike the loop above, which can quit when it drops a
1725 * delta, we need to keep going to look for more depth cuts. So we need
1726 * an extra "next" pointer to keep going after we reset cur->delta.
1728 for (cur
= entry
; cur
; cur
= next
) {
1732 * We should have a chain of zero or more ACTIVE states down to
1733 * a final DONE. We can quit after the DONE, because either it
1734 * has no bases, or we've already handled them in a previous
1737 if (cur
->dfs_state
== DFS_DONE
)
1739 else if (cur
->dfs_state
!= DFS_ACTIVE
)
1740 BUG("confusing delta dfs state in second pass: %d",
1744 * If the total_depth is more than depth, then we need to snip
1745 * the chain into two or more smaller chains that don't exceed
1746 * the maximum depth. Most of the resulting chains will contain
1747 * (depth + 1) entries (i.e., depth deltas plus one base), and
1748 * the last chain (i.e., the one containing entry) will contain
1749 * whatever entries are left over, namely
1750 * (total_depth % (depth + 1)) of them.
1752 * Since we are iterating towards decreasing depth, we need to
1753 * decrement total_depth as we go, and we need to write to the
1754 * entry what its final depth will be after all of the
1755 * snipping. Since we're snipping into chains of length (depth
1756 * + 1) entries, the final depth of an entry will be its
1757 * original depth modulo (depth + 1). Any time we encounter an
1758 * entry whose final depth is supposed to be zero, we snip it
1759 * from its delta base, thereby making it so.
1761 cur
->depth
= (total_depth
--) % (depth
+ 1);
1763 drop_reused_delta(cur
);
1765 cur
->dfs_state
= DFS_DONE
;
1769 static void get_object_details(void)
1772 struct object_entry
**sorted_by_offset
;
1775 progress_state
= start_progress(_("Counting objects"),
1776 to_pack
.nr_objects
);
1778 sorted_by_offset
= xcalloc(to_pack
.nr_objects
, sizeof(struct object_entry
*));
1779 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1780 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
1781 QSORT(sorted_by_offset
, to_pack
.nr_objects
, pack_offset_sort
);
1783 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
1784 struct object_entry
*entry
= sorted_by_offset
[i
];
1785 check_object(entry
);
1786 if (entry
->type_valid
&&
1787 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
))
1788 entry
->no_try_delta
= 1;
1789 display_progress(progress_state
, i
+ 1);
1791 stop_progress(&progress_state
);
1794 * This must happen in a second pass, since we rely on the delta
1795 * information for the whole list being completed.
1797 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1798 break_delta_chains(&to_pack
.objects
[i
]);
1800 free(sorted_by_offset
);
1804 * We search for deltas in a list sorted by type, by filename hash, and then
1805 * by size, so that we see progressively smaller and smaller files.
1806 * That's because we prefer deltas to be from the bigger file
1807 * to the smaller -- deletes are potentially cheaper, but perhaps
1808 * more importantly, the bigger file is likely the more recent
1809 * one. The deepest deltas are therefore the oldest objects which are
1810 * less susceptible to be accessed often.
1812 static int type_size_sort(const void *_a
, const void *_b
)
1814 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1815 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1816 enum object_type a_type
= oe_type(a
);
1817 enum object_type b_type
= oe_type(b
);
1818 unsigned long a_size
= SIZE(a
);
1819 unsigned long b_size
= SIZE(b
);
1821 if (a_type
> b_type
)
1823 if (a_type
< b_type
)
1825 if (a
->hash
> b
->hash
)
1827 if (a
->hash
< b
->hash
)
1829 if (a
->preferred_base
> b
->preferred_base
)
1831 if (a
->preferred_base
< b
->preferred_base
)
1833 if (a_size
> b_size
)
1835 if (a_size
< b_size
)
1837 return a
< b
? -1 : (a
> b
); /* newest first */
1841 struct object_entry
*entry
;
1843 struct delta_index
*index
;
1847 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1848 unsigned long delta_size
)
1850 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1853 if (delta_size
< cache_max_small_delta_size
)
1856 /* cache delta, if objects are large enough compared to delta size */
1857 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1865 static pthread_mutex_t read_mutex
;
1866 #define read_lock() pthread_mutex_lock(&read_mutex)
1867 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1869 static pthread_mutex_t cache_mutex
;
1870 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1871 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1873 static pthread_mutex_t progress_mutex
;
1874 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1875 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1879 #define read_lock() (void)0
1880 #define read_unlock() (void)0
1881 #define cache_lock() (void)0
1882 #define cache_unlock() (void)0
1883 #define progress_lock() (void)0
1884 #define progress_unlock() (void)0
1889 * Return the size of the object without doing any delta
1890 * reconstruction (so non-deltas are true object sizes, but deltas
1891 * return the size of the delta data).
1893 unsigned long oe_get_size_slow(struct packing_data
*pack
,
1894 const struct object_entry
*e
)
1896 struct packed_git
*p
;
1897 struct pack_window
*w_curs
;
1899 enum object_type type
;
1900 unsigned long used
, avail
, size
;
1902 if (e
->type_
!= OBJ_OFS_DELTA
&& e
->type_
!= OBJ_REF_DELTA
) {
1904 if (oid_object_info(the_repository
, &e
->idx
.oid
, &size
) < 0)
1905 die(_("unable to get size of %s"),
1906 oid_to_hex(&e
->idx
.oid
));
1911 p
= oe_in_pack(pack
, e
);
1913 BUG("when e->type is a delta, it must belong to a pack");
1917 buf
= use_pack(p
, &w_curs
, e
->in_pack_offset
, &avail
);
1918 used
= unpack_object_header_buffer(buf
, avail
, &type
, &size
);
1920 die(_("unable to parse object header of %s"),
1921 oid_to_hex(&e
->idx
.oid
));
1923 unuse_pack(&w_curs
);
1928 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1929 unsigned max_depth
, unsigned long *mem_usage
)
1931 struct object_entry
*trg_entry
= trg
->entry
;
1932 struct object_entry
*src_entry
= src
->entry
;
1933 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1935 enum object_type type
;
1938 /* Don't bother doing diffs between different types */
1939 if (oe_type(trg_entry
) != oe_type(src_entry
))
1943 * We do not bother to try a delta that we discarded on an
1944 * earlier try, but only when reusing delta data. Note that
1945 * src_entry that is marked as the preferred_base should always
1946 * be considered, as even if we produce a suboptimal delta against
1947 * it, we will still save the transfer cost, as we already know
1948 * the other side has it and we won't send src_entry at all.
1950 if (reuse_delta
&& IN_PACK(trg_entry
) &&
1951 IN_PACK(trg_entry
) == IN_PACK(src_entry
) &&
1952 !src_entry
->preferred_base
&&
1953 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1954 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1957 /* Let's not bust the allowed depth. */
1958 if (src
->depth
>= max_depth
)
1961 /* Now some size filtering heuristics. */
1962 trg_size
= SIZE(trg_entry
);
1963 if (!DELTA(trg_entry
)) {
1964 max_size
= trg_size
/2 - the_hash_algo
->rawsz
;
1967 max_size
= DELTA_SIZE(trg_entry
);
1968 ref_depth
= trg
->depth
;
1970 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
1971 (max_depth
- ref_depth
+ 1);
1974 src_size
= SIZE(src_entry
);
1975 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1976 if (sizediff
>= max_size
)
1978 if (trg_size
< src_size
/ 32)
1981 /* Load data if not already done */
1984 trg
->data
= read_object_file(&trg_entry
->idx
.oid
, &type
, &sz
);
1987 die("object %s cannot be read",
1988 oid_to_hex(&trg_entry
->idx
.oid
));
1990 die("object %s inconsistent object length (%lu vs %lu)",
1991 oid_to_hex(&trg_entry
->idx
.oid
), sz
,
1997 src
->data
= read_object_file(&src_entry
->idx
.oid
, &type
, &sz
);
2000 if (src_entry
->preferred_base
) {
2001 static int warned
= 0;
2003 warning("object %s cannot be read",
2004 oid_to_hex(&src_entry
->idx
.oid
));
2006 * Those objects are not included in the
2007 * resulting pack. Be resilient and ignore
2008 * them if they can't be read, in case the
2009 * pack could be created nevertheless.
2013 die("object %s cannot be read",
2014 oid_to_hex(&src_entry
->idx
.oid
));
2017 die("object %s inconsistent object length (%lu vs %lu)",
2018 oid_to_hex(&src_entry
->idx
.oid
), sz
,
2023 src
->index
= create_delta_index(src
->data
, src_size
);
2025 static int warned
= 0;
2027 warning("suboptimal pack - out of memory");
2030 *mem_usage
+= sizeof_delta_index(src
->index
);
2033 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
2036 if (delta_size
>= (1U << OE_DELTA_SIZE_BITS
)) {
2041 if (DELTA(trg_entry
)) {
2042 /* Prefer only shallower same-sized deltas. */
2043 if (delta_size
== DELTA_SIZE(trg_entry
) &&
2044 src
->depth
+ 1 >= trg
->depth
) {
2051 * Handle memory allocation outside of the cache
2052 * accounting lock. Compiler will optimize the strangeness
2053 * away when NO_PTHREADS is defined.
2055 free(trg_entry
->delta_data
);
2057 if (trg_entry
->delta_data
) {
2058 delta_cache_size
-= DELTA_SIZE(trg_entry
);
2059 trg_entry
->delta_data
= NULL
;
2061 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
2062 delta_cache_size
+= delta_size
;
2064 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
2070 SET_DELTA(trg_entry
, src_entry
);
2071 SET_DELTA_SIZE(trg_entry
, delta_size
);
2072 trg
->depth
= src
->depth
+ 1;
2077 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
2079 struct object_entry
*child
= DELTA_CHILD(me
);
2082 unsigned int c
= check_delta_limit(child
, n
+ 1);
2085 child
= DELTA_SIBLING(child
);
2090 static unsigned long free_unpacked(struct unpacked
*n
)
2092 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
2093 free_delta_index(n
->index
);
2096 freed_mem
+= SIZE(n
->entry
);
2097 FREE_AND_NULL(n
->data
);
2104 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
2105 int window
, int depth
, unsigned *processed
)
2107 uint32_t i
, idx
= 0, count
= 0;
2108 struct unpacked
*array
;
2109 unsigned long mem_usage
= 0;
2111 array
= xcalloc(window
, sizeof(struct unpacked
));
2114 struct object_entry
*entry
;
2115 struct unpacked
*n
= array
+ idx
;
2116 int j
, max_depth
, best_base
= -1;
2125 if (!entry
->preferred_base
) {
2127 display_progress(progress_state
, *processed
);
2131 mem_usage
-= free_unpacked(n
);
2134 while (window_memory_limit
&&
2135 mem_usage
> window_memory_limit
&&
2137 uint32_t tail
= (idx
+ window
- count
) % window
;
2138 mem_usage
-= free_unpacked(array
+ tail
);
2142 /* We do not compute delta to *create* objects we are not
2145 if (entry
->preferred_base
)
2149 * If the current object is at pack edge, take the depth the
2150 * objects that depend on the current object into account
2151 * otherwise they would become too deep.
2154 if (DELTA_CHILD(entry
)) {
2155 max_depth
-= check_delta_limit(entry
, 0);
2163 uint32_t other_idx
= idx
+ j
;
2165 if (other_idx
>= window
)
2166 other_idx
-= window
;
2167 m
= array
+ other_idx
;
2170 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
2174 best_base
= other_idx
;
2178 * If we decided to cache the delta data, then it is best
2179 * to compress it right away. First because we have to do
2180 * it anyway, and doing it here while we're threaded will
2181 * save a lot of time in the non threaded write phase,
2182 * as well as allow for caching more deltas within
2183 * the same cache size limit.
2185 * But only if not writing to stdout, since in that case
2186 * the network is most likely throttling writes anyway,
2187 * and therefore it is best to go to the write phase ASAP
2188 * instead, as we can afford spending more time compressing
2189 * between writes at that moment.
2191 if (entry
->delta_data
&& !pack_to_stdout
) {
2194 size
= do_compress(&entry
->delta_data
, DELTA_SIZE(entry
));
2195 if (size
< (1U << OE_Z_DELTA_BITS
)) {
2196 entry
->z_delta_size
= size
;
2198 delta_cache_size
-= DELTA_SIZE(entry
);
2199 delta_cache_size
+= entry
->z_delta_size
;
2202 FREE_AND_NULL(entry
->delta_data
);
2203 entry
->z_delta_size
= 0;
2207 /* if we made n a delta, and if n is already at max
2208 * depth, leaving it in the window is pointless. we
2209 * should evict it first.
2211 if (DELTA(entry
) && max_depth
<= n
->depth
)
2215 * Move the best delta base up in the window, after the
2216 * currently deltified object, to keep it longer. It will
2217 * be the first base object to be attempted next.
2220 struct unpacked swap
= array
[best_base
];
2221 int dist
= (window
+ idx
- best_base
) % window
;
2222 int dst
= best_base
;
2224 int src
= (dst
+ 1) % window
;
2225 array
[dst
] = array
[src
];
2233 if (count
+ 1 < window
)
2239 for (i
= 0; i
< window
; ++i
) {
2240 free_delta_index(array
[i
].index
);
2241 free(array
[i
].data
);
2248 static void try_to_free_from_threads(size_t size
)
2251 release_pack_memory(size
);
2255 static try_to_free_t old_try_to_free_routine
;
2258 * The main thread waits on the condition that (at least) one of the workers
2259 * has stopped working (which is indicated in the .working member of
2260 * struct thread_params).
2261 * When a work thread has completed its work, it sets .working to 0 and
2262 * signals the main thread and waits on the condition that .data_ready
2266 struct thread_params
{
2268 struct object_entry
**list
;
2275 pthread_mutex_t mutex
;
2276 pthread_cond_t cond
;
2277 unsigned *processed
;
2280 static pthread_cond_t progress_cond
;
2283 * Mutex and conditional variable can't be statically-initialized on Windows.
2285 static void init_threaded_search(void)
2287 init_recursive_mutex(&read_mutex
);
2288 pthread_mutex_init(&cache_mutex
, NULL
);
2289 pthread_mutex_init(&progress_mutex
, NULL
);
2290 pthread_cond_init(&progress_cond
, NULL
);
2291 old_try_to_free_routine
= set_try_to_free_routine(try_to_free_from_threads
);
2294 static void cleanup_threaded_search(void)
2296 set_try_to_free_routine(old_try_to_free_routine
);
2297 pthread_cond_destroy(&progress_cond
);
2298 pthread_mutex_destroy(&read_mutex
);
2299 pthread_mutex_destroy(&cache_mutex
);
2300 pthread_mutex_destroy(&progress_mutex
);
2303 static void *threaded_find_deltas(void *arg
)
2305 struct thread_params
*me
= arg
;
2308 while (me
->remaining
) {
2311 find_deltas(me
->list
, &me
->remaining
,
2312 me
->window
, me
->depth
, me
->processed
);
2316 pthread_cond_signal(&progress_cond
);
2320 * We must not set ->data_ready before we wait on the
2321 * condition because the main thread may have set it to 1
2322 * before we get here. In order to be sure that new
2323 * work is available if we see 1 in ->data_ready, it
2324 * was initialized to 0 before this thread was spawned
2325 * and we reset it to 0 right away.
2327 pthread_mutex_lock(&me
->mutex
);
2328 while (!me
->data_ready
)
2329 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2331 pthread_mutex_unlock(&me
->mutex
);
2336 /* leave ->working 1 so that this doesn't get more work assigned */
2340 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2341 int window
, int depth
, unsigned *processed
)
2343 struct thread_params
*p
;
2344 int i
, ret
, active_threads
= 0;
2346 init_threaded_search();
2348 if (delta_search_threads
<= 1) {
2349 find_deltas(list
, &list_size
, window
, depth
, processed
);
2350 cleanup_threaded_search();
2353 if (progress
> pack_to_stdout
)
2354 fprintf(stderr
, "Delta compression using up to %d threads.\n",
2355 delta_search_threads
);
2356 p
= xcalloc(delta_search_threads
, sizeof(*p
));
2358 /* Partition the work amongst work threads. */
2359 for (i
= 0; i
< delta_search_threads
; i
++) {
2360 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2362 /* don't use too small segments or no deltas will be found */
2363 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2366 p
[i
].window
= window
;
2368 p
[i
].processed
= processed
;
2370 p
[i
].data_ready
= 0;
2372 /* try to split chunks on "path" boundaries */
2373 while (sub_size
&& sub_size
< list_size
&&
2374 list
[sub_size
]->hash
&&
2375 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2379 p
[i
].list_size
= sub_size
;
2380 p
[i
].remaining
= sub_size
;
2383 list_size
-= sub_size
;
2386 /* Start work threads. */
2387 for (i
= 0; i
< delta_search_threads
; i
++) {
2388 if (!p
[i
].list_size
)
2390 pthread_mutex_init(&p
[i
].mutex
, NULL
);
2391 pthread_cond_init(&p
[i
].cond
, NULL
);
2392 ret
= pthread_create(&p
[i
].thread
, NULL
,
2393 threaded_find_deltas
, &p
[i
]);
2395 die("unable to create thread: %s", strerror(ret
));
2400 * Now let's wait for work completion. Each time a thread is done
2401 * with its work, we steal half of the remaining work from the
2402 * thread with the largest number of unprocessed objects and give
2403 * it to that newly idle thread. This ensure good load balancing
2404 * until the remaining object list segments are simply too short
2405 * to be worth splitting anymore.
2407 while (active_threads
) {
2408 struct thread_params
*target
= NULL
;
2409 struct thread_params
*victim
= NULL
;
2410 unsigned sub_size
= 0;
2414 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
2419 pthread_cond_wait(&progress_cond
, &progress_mutex
);
2422 for (i
= 0; i
< delta_search_threads
; i
++)
2423 if (p
[i
].remaining
> 2*window
&&
2424 (!victim
|| victim
->remaining
< p
[i
].remaining
))
2427 sub_size
= victim
->remaining
/ 2;
2428 list
= victim
->list
+ victim
->list_size
- sub_size
;
2429 while (sub_size
&& list
[0]->hash
&&
2430 list
[0]->hash
== list
[-1]->hash
) {
2436 * It is possible for some "paths" to have
2437 * so many objects that no hash boundary
2438 * might be found. Let's just steal the
2439 * exact half in that case.
2441 sub_size
= victim
->remaining
/ 2;
2444 target
->list
= list
;
2445 victim
->list_size
-= sub_size
;
2446 victim
->remaining
-= sub_size
;
2448 target
->list_size
= sub_size
;
2449 target
->remaining
= sub_size
;
2450 target
->working
= 1;
2453 pthread_mutex_lock(&target
->mutex
);
2454 target
->data_ready
= 1;
2455 pthread_cond_signal(&target
->cond
);
2456 pthread_mutex_unlock(&target
->mutex
);
2459 pthread_join(target
->thread
, NULL
);
2460 pthread_cond_destroy(&target
->cond
);
2461 pthread_mutex_destroy(&target
->mutex
);
2465 cleanup_threaded_search();
2470 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2473 static void add_tag_chain(const struct object_id
*oid
)
2478 * We catch duplicates already in add_object_entry(), but we'd
2479 * prefer to do this extra check to avoid having to parse the
2480 * tag at all if we already know that it's being packed (e.g., if
2481 * it was included via bitmaps, we would not have parsed it
2484 if (packlist_find(&to_pack
, oid
->hash
, NULL
))
2487 tag
= lookup_tag(the_repository
, oid
);
2489 if (!tag
|| parse_tag(tag
) || !tag
->tagged
)
2490 die("unable to pack objects reachable from tag %s",
2493 add_object_entry(&tag
->object
.oid
, OBJ_TAG
, NULL
, 0);
2495 if (tag
->tagged
->type
!= OBJ_TAG
)
2498 tag
= (struct tag
*)tag
->tagged
;
2502 static int add_ref_tag(const char *path
, const struct object_id
*oid
, int flag
, void *cb_data
)
2504 struct object_id peeled
;
2506 if (starts_with(path
, "refs/tags/") && /* is a tag? */
2507 !peel_ref(path
, &peeled
) && /* peelable? */
2508 packlist_find(&to_pack
, peeled
.hash
, NULL
)) /* object packed? */
2513 static void prepare_pack(int window
, int depth
)
2515 struct object_entry
**delta_list
;
2516 uint32_t i
, nr_deltas
;
2519 get_object_details();
2522 * If we're locally repacking then we need to be doubly careful
2523 * from now on in order to make sure no stealth corruption gets
2524 * propagated to the new pack. Clients receiving streamed packs
2525 * should validate everything they get anyway so no need to incur
2526 * the additional cost here in that case.
2528 if (!pack_to_stdout
)
2529 do_check_packed_object_crc
= 1;
2531 if (!to_pack
.nr_objects
|| !window
|| !depth
)
2534 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
2537 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2538 struct object_entry
*entry
= to_pack
.objects
+ i
;
2541 /* This happens if we decided to reuse existing
2542 * delta from a pack. "reuse_delta &&" is implied.
2546 if (!entry
->type_valid
||
2547 oe_size_less_than(&to_pack
, entry
, 50))
2550 if (entry
->no_try_delta
)
2553 if (!entry
->preferred_base
) {
2555 if (oe_type(entry
) < 0)
2556 die("unable to get type of object %s",
2557 oid_to_hex(&entry
->idx
.oid
));
2559 if (oe_type(entry
) < 0) {
2561 * This object is not found, but we
2562 * don't have to include it anyway.
2568 delta_list
[n
++] = entry
;
2571 if (nr_deltas
&& n
> 1) {
2572 unsigned nr_done
= 0;
2574 progress_state
= start_progress(_("Compressing objects"),
2576 QSORT(delta_list
, n
, type_size_sort
);
2577 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
2578 stop_progress(&progress_state
);
2579 if (nr_done
!= nr_deltas
)
2580 die("inconsistency with delta count");
2585 static int git_pack_config(const char *k
, const char *v
, void *cb
)
2587 if (!strcmp(k
, "pack.window")) {
2588 window
= git_config_int(k
, v
);
2591 if (!strcmp(k
, "pack.windowmemory")) {
2592 window_memory_limit
= git_config_ulong(k
, v
);
2595 if (!strcmp(k
, "pack.depth")) {
2596 depth
= git_config_int(k
, v
);
2599 if (!strcmp(k
, "pack.deltacachesize")) {
2600 max_delta_cache_size
= git_config_int(k
, v
);
2603 if (!strcmp(k
, "pack.deltacachelimit")) {
2604 cache_max_small_delta_size
= git_config_int(k
, v
);
2607 if (!strcmp(k
, "pack.writebitmaphashcache")) {
2608 if (git_config_bool(k
, v
))
2609 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
2611 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
2613 if (!strcmp(k
, "pack.usebitmaps")) {
2614 use_bitmap_index_default
= git_config_bool(k
, v
);
2617 if (!strcmp(k
, "pack.threads")) {
2618 delta_search_threads
= git_config_int(k
, v
);
2619 if (delta_search_threads
< 0)
2620 die("invalid number of threads specified (%d)",
2621 delta_search_threads
);
2623 if (delta_search_threads
!= 1) {
2624 warning("no threads support, ignoring %s", k
);
2625 delta_search_threads
= 0;
2630 if (!strcmp(k
, "pack.indexversion")) {
2631 pack_idx_opts
.version
= git_config_int(k
, v
);
2632 if (pack_idx_opts
.version
> 2)
2633 die("bad pack.indexversion=%"PRIu32
,
2634 pack_idx_opts
.version
);
2637 return git_default_config(k
, v
, cb
);
2640 static void read_object_list_from_stdin(void)
2642 char line
[GIT_MAX_HEXSZ
+ 1 + PATH_MAX
+ 2];
2643 struct object_id oid
;
2647 if (!fgets(line
, sizeof(line
), stdin
)) {
2651 die("fgets returned NULL, not EOF, not error!");
2657 if (line
[0] == '-') {
2658 if (get_oid_hex(line
+1, &oid
))
2659 die("expected edge object ID, got garbage:\n %s",
2661 add_preferred_base(&oid
);
2664 if (parse_oid_hex(line
, &oid
, &p
))
2665 die("expected object ID, got garbage:\n %s", line
);
2667 add_preferred_base_object(p
+ 1);
2668 add_object_entry(&oid
, OBJ_NONE
, p
+ 1, 0);
2672 /* Remember to update object flag allocation in object.h */
2673 #define OBJECT_ADDED (1u<<20)
2675 static void show_commit(struct commit
*commit
, void *data
)
2677 add_object_entry(&commit
->object
.oid
, OBJ_COMMIT
, NULL
, 0);
2678 commit
->object
.flags
|= OBJECT_ADDED
;
2680 if (write_bitmap_index
)
2681 index_commit_for_bitmap(commit
);
2684 static void show_object(struct object
*obj
, const char *name
, void *data
)
2686 add_preferred_base_object(name
);
2687 add_object_entry(&obj
->oid
, obj
->type
, name
, 0);
2688 obj
->flags
|= OBJECT_ADDED
;
2691 static void show_object__ma_allow_any(struct object
*obj
, const char *name
, void *data
)
2693 assert(arg_missing_action
== MA_ALLOW_ANY
);
2696 * Quietly ignore ALL missing objects. This avoids problems with
2697 * staging them now and getting an odd error later.
2699 if (!has_object_file(&obj
->oid
))
2702 show_object(obj
, name
, data
);
2705 static void show_object__ma_allow_promisor(struct object
*obj
, const char *name
, void *data
)
2707 assert(arg_missing_action
== MA_ALLOW_PROMISOR
);
2710 * Quietly ignore EXPECTED missing objects. This avoids problems with
2711 * staging them now and getting an odd error later.
2713 if (!has_object_file(&obj
->oid
) && is_promisor_object(&obj
->oid
))
2716 show_object(obj
, name
, data
);
2719 static int option_parse_missing_action(const struct option
*opt
,
2720 const char *arg
, int unset
)
2725 if (!strcmp(arg
, "error")) {
2726 arg_missing_action
= MA_ERROR
;
2727 fn_show_object
= show_object
;
2731 if (!strcmp(arg
, "allow-any")) {
2732 arg_missing_action
= MA_ALLOW_ANY
;
2733 fetch_if_missing
= 0;
2734 fn_show_object
= show_object__ma_allow_any
;
2738 if (!strcmp(arg
, "allow-promisor")) {
2739 arg_missing_action
= MA_ALLOW_PROMISOR
;
2740 fetch_if_missing
= 0;
2741 fn_show_object
= show_object__ma_allow_promisor
;
2745 die(_("invalid value for --missing"));
2749 static void show_edge(struct commit
*commit
)
2751 add_preferred_base(&commit
->object
.oid
);
2754 struct in_pack_object
{
2756 struct object
*object
;
2762 struct in_pack_object
*array
;
2765 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
2767 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->oid
.hash
, p
);
2768 in_pack
->array
[in_pack
->nr
].object
= object
;
2773 * Compare the objects in the offset order, in order to emulate the
2774 * "git rev-list --objects" output that produced the pack originally.
2776 static int ofscmp(const void *a_
, const void *b_
)
2778 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
2779 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
2781 if (a
->offset
< b
->offset
)
2783 else if (a
->offset
> b
->offset
)
2786 return oidcmp(&a
->object
->oid
, &b
->object
->oid
);
2789 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
2791 struct packed_git
*p
;
2792 struct in_pack in_pack
;
2795 memset(&in_pack
, 0, sizeof(in_pack
));
2797 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
2798 struct object_id oid
;
2801 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
2803 if (open_pack_index(p
))
2804 die("cannot open pack index");
2806 ALLOC_GROW(in_pack
.array
,
2807 in_pack
.nr
+ p
->num_objects
,
2810 for (i
= 0; i
< p
->num_objects
; i
++) {
2811 nth_packed_object_oid(&oid
, p
, i
);
2812 o
= lookup_unknown_object(oid
.hash
);
2813 if (!(o
->flags
& OBJECT_ADDED
))
2814 mark_in_pack_object(o
, p
, &in_pack
);
2815 o
->flags
|= OBJECT_ADDED
;
2820 QSORT(in_pack
.array
, in_pack
.nr
, ofscmp
);
2821 for (i
= 0; i
< in_pack
.nr
; i
++) {
2822 struct object
*o
= in_pack
.array
[i
].object
;
2823 add_object_entry(&o
->oid
, o
->type
, "", 0);
2826 free(in_pack
.array
);
2829 static int add_loose_object(const struct object_id
*oid
, const char *path
,
2832 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
2835 warning("loose object at %s could not be examined", path
);
2839 add_object_entry(oid
, type
, "", 0);
2844 * We actually don't even have to worry about reachability here.
2845 * add_object_entry will weed out duplicates, so we just add every
2846 * loose object we find.
2848 static void add_unreachable_loose_objects(void)
2850 for_each_loose_file_in_objdir(get_object_directory(),
2855 static int has_sha1_pack_kept_or_nonlocal(const struct object_id
*oid
)
2857 static struct packed_git
*last_found
= (void *)1;
2858 struct packed_git
*p
;
2860 p
= (last_found
!= (void *)1) ? last_found
:
2861 get_packed_git(the_repository
);
2864 if ((!p
->pack_local
|| p
->pack_keep
||
2865 p
->pack_keep_in_core
) &&
2866 find_pack_entry_one(oid
->hash
, p
)) {
2870 if (p
== last_found
)
2871 p
= get_packed_git(the_repository
);
2874 if (p
== last_found
)
2881 * Store a list of sha1s that are should not be discarded
2882 * because they are either written too recently, or are
2883 * reachable from another object that was.
2885 * This is filled by get_object_list.
2887 static struct oid_array recent_objects
;
2889 static int loosened_object_can_be_discarded(const struct object_id
*oid
,
2892 if (!unpack_unreachable_expiration
)
2894 if (mtime
> unpack_unreachable_expiration
)
2896 if (oid_array_lookup(&recent_objects
, oid
) >= 0)
2901 static void loosen_unused_packed_objects(struct rev_info
*revs
)
2903 struct packed_git
*p
;
2905 struct object_id oid
;
2907 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
2908 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
2911 if (open_pack_index(p
))
2912 die("cannot open pack index");
2914 for (i
= 0; i
< p
->num_objects
; i
++) {
2915 nth_packed_object_oid(&oid
, p
, i
);
2916 if (!packlist_find(&to_pack
, oid
.hash
, NULL
) &&
2917 !has_sha1_pack_kept_or_nonlocal(&oid
) &&
2918 !loosened_object_can_be_discarded(&oid
, p
->mtime
))
2919 if (force_object_loose(&oid
, p
->mtime
))
2920 die("unable to force loose object");
2926 * This tracks any options which pack-reuse code expects to be on, or which a
2927 * reader of the pack might not understand, and which would therefore prevent
2928 * blind reuse of what we have on disk.
2930 static int pack_options_allow_reuse(void)
2932 return pack_to_stdout
&&
2934 !ignore_packed_keep_on_disk
&&
2935 !ignore_packed_keep_in_core
&&
2936 (!local
|| !have_non_local_packs
) &&
2940 static int get_object_list_from_bitmap(struct rev_info
*revs
)
2942 struct bitmap_index
*bitmap_git
;
2943 if (!(bitmap_git
= prepare_bitmap_walk(revs
)))
2946 if (pack_options_allow_reuse() &&
2947 !reuse_partial_packfile_from_bitmap(
2950 &reuse_packfile_objects
,
2951 &reuse_packfile_offset
)) {
2952 assert(reuse_packfile_objects
);
2953 nr_result
+= reuse_packfile_objects
;
2954 display_progress(progress_state
, nr_result
);
2957 traverse_bitmap_commit_list(bitmap_git
, &add_object_entry_from_bitmap
);
2958 free_bitmap_index(bitmap_git
);
2962 static void record_recent_object(struct object
*obj
,
2966 oid_array_append(&recent_objects
, &obj
->oid
);
2969 static void record_recent_commit(struct commit
*commit
, void *data
)
2971 oid_array_append(&recent_objects
, &commit
->object
.oid
);
2974 static void get_object_list(int ac
, const char **av
)
2976 struct rev_info revs
;
2980 init_revisions(&revs
, NULL
);
2981 save_commit_buffer
= 0;
2982 setup_revisions(ac
, av
, &revs
, NULL
);
2984 /* make sure shallows are read */
2985 is_repository_shallow(the_repository
);
2987 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2988 int len
= strlen(line
);
2989 if (len
&& line
[len
- 1] == '\n')
2994 if (!strcmp(line
, "--not")) {
2995 flags
^= UNINTERESTING
;
2996 write_bitmap_index
= 0;
2999 if (starts_with(line
, "--shallow ")) {
3000 struct object_id oid
;
3001 if (get_oid_hex(line
+ 10, &oid
))
3002 die("not an SHA-1 '%s'", line
+ 10);
3003 register_shallow(the_repository
, &oid
);
3004 use_bitmap_index
= 0;
3007 die("not a rev '%s'", line
);
3009 if (handle_revision_arg(line
, &revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
3010 die("bad revision '%s'", line
);
3013 if (use_bitmap_index
&& !get_object_list_from_bitmap(&revs
))
3016 if (prepare_revision_walk(&revs
))
3017 die("revision walk setup failed");
3018 mark_edges_uninteresting(&revs
, show_edge
);
3020 if (!fn_show_object
)
3021 fn_show_object
= show_object
;
3022 traverse_commit_list_filtered(&filter_options
, &revs
,
3023 show_commit
, fn_show_object
, NULL
,
3026 if (unpack_unreachable_expiration
) {
3027 revs
.ignore_missing_links
= 1;
3028 if (add_unseen_recent_objects_to_traversal(&revs
,
3029 unpack_unreachable_expiration
))
3030 die("unable to add recent objects");
3031 if (prepare_revision_walk(&revs
))
3032 die("revision walk setup failed");
3033 traverse_commit_list(&revs
, record_recent_commit
,
3034 record_recent_object
, NULL
);
3037 if (keep_unreachable
)
3038 add_objects_in_unpacked_packs(&revs
);
3039 if (pack_loose_unreachable
)
3040 add_unreachable_loose_objects();
3041 if (unpack_unreachable
)
3042 loosen_unused_packed_objects(&revs
);
3044 oid_array_clear(&recent_objects
);
3047 static void add_extra_kept_packs(const struct string_list
*names
)
3049 struct packed_git
*p
;
3054 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
3055 const char *name
= basename(p
->pack_name
);
3061 for (i
= 0; i
< names
->nr
; i
++)
3062 if (!fspathcmp(name
, names
->items
[i
].string
))
3065 if (i
< names
->nr
) {
3066 p
->pack_keep_in_core
= 1;
3067 ignore_packed_keep_in_core
= 1;
3073 static int option_parse_index_version(const struct option
*opt
,
3074 const char *arg
, int unset
)
3077 const char *val
= arg
;
3078 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
3079 if (pack_idx_opts
.version
> 2)
3080 die(_("unsupported index version %s"), val
);
3081 if (*c
== ',' && c
[1])
3082 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
3083 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
3084 die(_("bad index version '%s'"), val
);
3088 static int option_parse_unpack_unreachable(const struct option
*opt
,
3089 const char *arg
, int unset
)
3092 unpack_unreachable
= 0;
3093 unpack_unreachable_expiration
= 0;
3096 unpack_unreachable
= 1;
3098 unpack_unreachable_expiration
= approxidate(arg
);
3103 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
3105 int use_internal_rev_list
= 0;
3108 int all_progress_implied
= 0;
3109 struct argv_array rp
= ARGV_ARRAY_INIT
;
3110 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
3111 int rev_list_index
= 0;
3112 struct string_list keep_pack_list
= STRING_LIST_INIT_NODUP
;
3113 struct option pack_objects_options
[] = {
3114 OPT_SET_INT('q', "quiet", &progress
,
3115 N_("do not show progress meter"), 0),
3116 OPT_SET_INT(0, "progress", &progress
,
3117 N_("show progress meter"), 1),
3118 OPT_SET_INT(0, "all-progress", &progress
,
3119 N_("show progress meter during object writing phase"), 2),
3120 OPT_BOOL(0, "all-progress-implied",
3121 &all_progress_implied
,
3122 N_("similar to --all-progress when progress meter is shown")),
3123 { OPTION_CALLBACK
, 0, "index-version", NULL
, N_("version[,offset]"),
3124 N_("write the pack index file in the specified idx format version"),
3125 0, option_parse_index_version
},
3126 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
3127 N_("maximum size of each output pack file")),
3128 OPT_BOOL(0, "local", &local
,
3129 N_("ignore borrowed objects from alternate object store")),
3130 OPT_BOOL(0, "incremental", &incremental
,
3131 N_("ignore packed objects")),
3132 OPT_INTEGER(0, "window", &window
,
3133 N_("limit pack window by objects")),
3134 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
3135 N_("limit pack window by memory in addition to object limit")),
3136 OPT_INTEGER(0, "depth", &depth
,
3137 N_("maximum length of delta chain allowed in the resulting pack")),
3138 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
3139 N_("reuse existing deltas")),
3140 OPT_BOOL(0, "reuse-object", &reuse_object
,
3141 N_("reuse existing objects")),
3142 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
3143 N_("use OFS_DELTA objects")),
3144 OPT_INTEGER(0, "threads", &delta_search_threads
,
3145 N_("use threads when searching for best delta matches")),
3146 OPT_BOOL(0, "non-empty", &non_empty
,
3147 N_("do not create an empty pack output")),
3148 OPT_BOOL(0, "revs", &use_internal_rev_list
,
3149 N_("read revision arguments from standard input")),
3150 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked
,
3151 N_("limit the objects to those that are not yet packed"),
3152 1, PARSE_OPT_NONEG
),
3153 OPT_SET_INT_F(0, "all", &rev_list_all
,
3154 N_("include objects reachable from any reference"),
3155 1, PARSE_OPT_NONEG
),
3156 OPT_SET_INT_F(0, "reflog", &rev_list_reflog
,
3157 N_("include objects referred by reflog entries"),
3158 1, PARSE_OPT_NONEG
),
3159 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index
,
3160 N_("include objects referred to by the index"),
3161 1, PARSE_OPT_NONEG
),
3162 OPT_BOOL(0, "stdout", &pack_to_stdout
,
3163 N_("output pack to stdout")),
3164 OPT_BOOL(0, "include-tag", &include_tag
,
3165 N_("include tag objects that refer to objects to be packed")),
3166 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
3167 N_("keep unreachable objects")),
3168 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
3169 N_("pack loose unreachable objects")),
3170 { OPTION_CALLBACK
, 0, "unpack-unreachable", NULL
, N_("time"),
3171 N_("unpack unreachable objects newer than <time>"),
3172 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
},
3173 OPT_BOOL(0, "thin", &thin
,
3174 N_("create thin packs")),
3175 OPT_BOOL(0, "shallow", &shallow
,
3176 N_("create packs suitable for shallow fetches")),
3177 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk
,
3178 N_("ignore packs that have companion .keep file")),
3179 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list
, N_("name"),
3180 N_("ignore this pack")),
3181 OPT_INTEGER(0, "compression", &pack_compression_level
,
3182 N_("pack compression level")),
3183 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
3184 N_("do not hide commits by grafts"), 0),
3185 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
3186 N_("use a bitmap index if available to speed up counting objects")),
3187 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index
,
3188 N_("write a bitmap index together with the pack index")),
3189 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options
),
3190 { OPTION_CALLBACK
, 0, "missing", NULL
, N_("action"),
3191 N_("handling for missing objects"), PARSE_OPT_NONEG
,
3192 option_parse_missing_action
},
3193 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects
,
3194 N_("do not pack objects in promisor packfiles")),
3198 if (DFS_NUM_STATES
> (1 << OE_DFS_STATE_BITS
))
3199 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3201 check_replace_refs
= 0;
3203 reset_pack_idx_option(&pack_idx_opts
);
3204 git_config(git_pack_config
, NULL
);
3206 progress
= isatty(2);
3207 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
3211 base_name
= argv
[0];
3214 if (pack_to_stdout
!= !base_name
|| argc
)
3215 usage_with_options(pack_usage
, pack_objects_options
);
3217 if (depth
>= (1 << OE_DEPTH_BITS
)) {
3218 warning(_("delta chain depth %d is too deep, forcing %d"),
3219 depth
, (1 << OE_DEPTH_BITS
) - 1);
3220 depth
= (1 << OE_DEPTH_BITS
) - 1;
3222 if (cache_max_small_delta_size
>= (1U << OE_Z_DELTA_BITS
)) {
3223 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3224 (1U << OE_Z_DELTA_BITS
) - 1);
3225 cache_max_small_delta_size
= (1U << OE_Z_DELTA_BITS
) - 1;
3228 argv_array_push(&rp
, "pack-objects");
3230 use_internal_rev_list
= 1;
3231 argv_array_push(&rp
, shallow
3232 ? "--objects-edge-aggressive"
3233 : "--objects-edge");
3235 argv_array_push(&rp
, "--objects");
3238 use_internal_rev_list
= 1;
3239 argv_array_push(&rp
, "--all");
3241 if (rev_list_reflog
) {
3242 use_internal_rev_list
= 1;
3243 argv_array_push(&rp
, "--reflog");
3245 if (rev_list_index
) {
3246 use_internal_rev_list
= 1;
3247 argv_array_push(&rp
, "--indexed-objects");
3249 if (rev_list_unpacked
) {
3250 use_internal_rev_list
= 1;
3251 argv_array_push(&rp
, "--unpacked");
3254 if (exclude_promisor_objects
) {
3255 use_internal_rev_list
= 1;
3256 fetch_if_missing
= 0;
3257 argv_array_push(&rp
, "--exclude-promisor-objects");
3259 if (unpack_unreachable
|| keep_unreachable
|| pack_loose_unreachable
)
3260 use_internal_rev_list
= 1;
3264 if (pack_compression_level
== -1)
3265 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
3266 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
3267 die("bad pack compression level %d", pack_compression_level
);
3269 if (!delta_search_threads
) /* --threads=0 means autodetect */
3270 delta_search_threads
= online_cpus();
3273 if (delta_search_threads
!= 1)
3274 warning("no threads support, ignoring --threads");
3276 if (!pack_to_stdout
&& !pack_size_limit
)
3277 pack_size_limit
= pack_size_limit_cfg
;
3278 if (pack_to_stdout
&& pack_size_limit
)
3279 die("--max-pack-size cannot be used to build a pack for transfer.");
3280 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
3281 warning("minimum pack size limit is 1 MiB");
3282 pack_size_limit
= 1024*1024;
3285 if (!pack_to_stdout
&& thin
)
3286 die("--thin cannot be used to build an indexable pack.");
3288 if (keep_unreachable
&& unpack_unreachable
)
3289 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3290 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
3291 unpack_unreachable_expiration
= 0;
3293 if (filter_options
.choice
) {
3294 if (!pack_to_stdout
)
3295 die("cannot use --filter without --stdout.");
3296 use_bitmap_index
= 0;
3300 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3302 * - to produce good pack (with bitmap index not-yet-packed objects are
3303 * packed in suboptimal order).
3305 * - to use more robust pack-generation codepath (avoiding possible
3306 * bugs in bitmap code and possible bitmap index corruption).
3308 if (!pack_to_stdout
)
3309 use_bitmap_index_default
= 0;
3311 if (use_bitmap_index
< 0)
3312 use_bitmap_index
= use_bitmap_index_default
;
3314 /* "hard" reasons not to use bitmaps; these just won't work at all */
3315 if (!use_internal_rev_list
|| (!pack_to_stdout
&& write_bitmap_index
) || is_repository_shallow(the_repository
))
3316 use_bitmap_index
= 0;
3318 if (pack_to_stdout
|| !rev_list_all
)
3319 write_bitmap_index
= 0;
3321 if (progress
&& all_progress_implied
)
3324 add_extra_kept_packs(&keep_pack_list
);
3325 if (ignore_packed_keep_on_disk
) {
3326 struct packed_git
*p
;
3327 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
)
3328 if (p
->pack_local
&& p
->pack_keep
)
3330 if (!p
) /* no keep-able packs found */
3331 ignore_packed_keep_on_disk
= 0;
3335 * unlike ignore_packed_keep_on_disk above, we do not
3336 * want to unset "local" based on looking at packs, as
3337 * it also covers non-local objects
3339 struct packed_git
*p
;
3340 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
3341 if (!p
->pack_local
) {
3342 have_non_local_packs
= 1;
3348 prepare_packing_data(&to_pack
);
3351 progress_state
= start_progress(_("Enumerating objects"), 0);
3352 if (!use_internal_rev_list
)
3353 read_object_list_from_stdin();
3355 get_object_list(rp
.argc
, rp
.argv
);
3356 argv_array_clear(&rp
);
3358 cleanup_preferred_base();
3359 if (include_tag
&& nr_result
)
3360 for_each_ref(add_ref_tag
, NULL
);
3361 stop_progress(&progress_state
);
3363 if (non_empty
&& !nr_result
)
3366 prepare_pack(window
, depth
);
3369 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
3370 " reused %"PRIu32
" (delta %"PRIu32
")\n",
3371 written
, written_delta
, reused
, reused_delta
);