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 "delta-islands.h"
28 #include "reachable.h"
29 #include "sha1-array.h"
30 #include "argv-array.h"
33 #include "object-store.h"
36 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
37 #define SIZE(obj) oe_size(&to_pack, obj)
38 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
39 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
40 #define DELTA(obj) oe_delta(&to_pack, obj)
41 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
42 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
43 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
44 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
45 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
46 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
48 static const char *pack_usage
[] = {
49 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
50 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
55 * Objects we are going to pack are collected in the `to_pack` structure.
56 * It contains an array (dynamically expanded) of the object data, and a map
57 * that can resolve SHA1s to their position in the array.
59 static struct packing_data to_pack
;
61 static struct pack_idx_entry
**written_list
;
62 static uint32_t nr_result
, nr_written
, nr_seen
;
63 static uint32_t write_layer
;
66 static int reuse_delta
= 1, reuse_object
= 1;
67 static int keep_unreachable
, unpack_unreachable
, include_tag
;
68 static timestamp_t unpack_unreachable_expiration
;
69 static int pack_loose_unreachable
;
71 static int have_non_local_packs
;
72 static int incremental
;
73 static int ignore_packed_keep_on_disk
;
74 static int ignore_packed_keep_in_core
;
75 static int allow_ofs_delta
;
76 static struct pack_idx_option pack_idx_opts
;
77 static const char *base_name
;
78 static int progress
= 1;
79 static int window
= 10;
80 static unsigned long pack_size_limit
;
81 static int depth
= 50;
82 static int delta_search_threads
;
83 static int pack_to_stdout
;
84 static int num_preferred_base
;
85 static struct progress
*progress_state
;
87 static struct packed_git
*reuse_packfile
;
88 static uint32_t reuse_packfile_objects
;
89 static off_t reuse_packfile_offset
;
91 static int use_bitmap_index_default
= 1;
92 static int use_bitmap_index
= -1;
93 static int write_bitmap_index
;
94 static uint16_t write_bitmap_options
;
96 static int exclude_promisor_objects
;
98 static int use_delta_islands
;
100 static unsigned long delta_cache_size
= 0;
101 static unsigned long max_delta_cache_size
= DEFAULT_DELTA_CACHE_SIZE
;
102 static unsigned long cache_max_small_delta_size
= 1000;
104 static unsigned long window_memory_limit
= 0;
106 static struct list_objects_filter_options filter_options
;
108 enum missing_action
{
109 MA_ERROR
= 0, /* fail if any missing objects are encountered */
110 MA_ALLOW_ANY
, /* silently allow ALL missing objects */
111 MA_ALLOW_PROMISOR
, /* silently allow all missing PROMISOR objects */
113 static enum missing_action arg_missing_action
;
114 static show_object_fn fn_show_object
;
119 static uint32_t written
, written_delta
;
120 static uint32_t reused
, reused_delta
;
125 static struct commit
**indexed_commits
;
126 static unsigned int indexed_commits_nr
;
127 static unsigned int indexed_commits_alloc
;
129 static void index_commit_for_bitmap(struct commit
*commit
)
131 if (indexed_commits_nr
>= indexed_commits_alloc
) {
132 indexed_commits_alloc
= (indexed_commits_alloc
+ 32) * 2;
133 REALLOC_ARRAY(indexed_commits
, indexed_commits_alloc
);
136 indexed_commits
[indexed_commits_nr
++] = commit
;
139 static void *get_delta(struct object_entry
*entry
)
141 unsigned long size
, base_size
, delta_size
;
142 void *buf
, *base_buf
, *delta_buf
;
143 enum object_type type
;
145 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
147 die("unable to read %s", oid_to_hex(&entry
->idx
.oid
));
148 base_buf
= read_object_file(&DELTA(entry
)->idx
.oid
, &type
,
151 die("unable to read %s",
152 oid_to_hex(&DELTA(entry
)->idx
.oid
));
153 delta_buf
= diff_delta(base_buf
, base_size
,
154 buf
, size
, &delta_size
, 0);
155 if (!delta_buf
|| delta_size
!= DELTA_SIZE(entry
))
156 die("delta size changed");
162 static unsigned long do_compress(void **pptr
, unsigned long size
)
166 unsigned long maxsize
;
168 git_deflate_init(&stream
, pack_compression_level
);
169 maxsize
= git_deflate_bound(&stream
, size
);
172 out
= xmalloc(maxsize
);
176 stream
.avail_in
= size
;
177 stream
.next_out
= out
;
178 stream
.avail_out
= maxsize
;
179 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
181 git_deflate_end(&stream
);
184 return stream
.total_out
;
187 static unsigned long write_large_blob_data(struct git_istream
*st
, struct hashfile
*f
,
188 const struct object_id
*oid
)
191 unsigned char ibuf
[1024 * 16];
192 unsigned char obuf
[1024 * 16];
193 unsigned long olen
= 0;
195 git_deflate_init(&stream
, pack_compression_level
);
200 readlen
= read_istream(st
, ibuf
, sizeof(ibuf
));
202 die(_("unable to read %s"), oid_to_hex(oid
));
204 stream
.next_in
= ibuf
;
205 stream
.avail_in
= readlen
;
206 while ((stream
.avail_in
|| readlen
== 0) &&
207 (zret
== Z_OK
|| zret
== Z_BUF_ERROR
)) {
208 stream
.next_out
= obuf
;
209 stream
.avail_out
= sizeof(obuf
);
210 zret
= git_deflate(&stream
, readlen
? 0 : Z_FINISH
);
211 hashwrite(f
, obuf
, stream
.next_out
- obuf
);
212 olen
+= stream
.next_out
- obuf
;
215 die(_("deflate error (%d)"), zret
);
217 if (zret
!= Z_STREAM_END
)
218 die(_("deflate error (%d)"), zret
);
222 git_deflate_end(&stream
);
227 * we are going to reuse the existing object data as is. make
228 * sure it is not corrupt.
230 static int check_pack_inflate(struct packed_git
*p
,
231 struct pack_window
**w_curs
,
234 unsigned long expect
)
237 unsigned char fakebuf
[4096], *in
;
240 memset(&stream
, 0, sizeof(stream
));
241 git_inflate_init(&stream
);
243 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
245 stream
.next_out
= fakebuf
;
246 stream
.avail_out
= sizeof(fakebuf
);
247 st
= git_inflate(&stream
, Z_FINISH
);
248 offset
+= stream
.next_in
- in
;
249 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
250 git_inflate_end(&stream
);
251 return (st
== Z_STREAM_END
&&
252 stream
.total_out
== expect
&&
253 stream
.total_in
== len
) ? 0 : -1;
256 static void copy_pack_data(struct hashfile
*f
,
257 struct packed_git
*p
,
258 struct pack_window
**w_curs
,
266 in
= use_pack(p
, w_curs
, offset
, &avail
);
268 avail
= (unsigned long)len
;
269 hashwrite(f
, in
, avail
);
275 /* Return 0 if we will bust the pack-size limit */
276 static unsigned long write_no_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
277 unsigned long limit
, int usable_delta
)
279 unsigned long size
, datalen
;
280 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
281 dheader
[MAX_PACK_OBJECT_HEADER
];
283 enum object_type type
;
285 struct git_istream
*st
= NULL
;
286 const unsigned hashsz
= the_hash_algo
->rawsz
;
289 if (oe_type(entry
) == OBJ_BLOB
&&
290 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
) &&
291 (st
= open_istream(&entry
->idx
.oid
, &type
, &size
, NULL
)) != NULL
)
294 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
296 die(_("unable to read %s"),
297 oid_to_hex(&entry
->idx
.oid
));
300 * make sure no cached delta data remains from a
301 * previous attempt before a pack split occurred.
303 FREE_AND_NULL(entry
->delta_data
);
304 entry
->z_delta_size
= 0;
305 } else if (entry
->delta_data
) {
306 size
= DELTA_SIZE(entry
);
307 buf
= entry
->delta_data
;
308 entry
->delta_data
= NULL
;
309 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
310 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
312 buf
= get_delta(entry
);
313 size
= DELTA_SIZE(entry
);
314 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
315 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
318 if (st
) /* large blob case, just assume we don't compress well */
320 else if (entry
->z_delta_size
)
321 datalen
= entry
->z_delta_size
;
323 datalen
= do_compress(&buf
, size
);
326 * The object header is a byte of 'type' followed by zero or
327 * more bytes of length.
329 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
332 if (type
== OBJ_OFS_DELTA
) {
334 * Deltas with relative base contain an additional
335 * encoding of the relative offset for the delta
336 * base from this object's position in the pack.
338 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
339 unsigned pos
= sizeof(dheader
) - 1;
340 dheader
[pos
] = ofs
& 127;
342 dheader
[--pos
] = 128 | (--ofs
& 127);
343 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
349 hashwrite(f
, header
, hdrlen
);
350 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
351 hdrlen
+= sizeof(dheader
) - pos
;
352 } else if (type
== OBJ_REF_DELTA
) {
354 * Deltas with a base reference contain
355 * additional bytes for the base object ID.
357 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
363 hashwrite(f
, header
, hdrlen
);
364 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
367 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
373 hashwrite(f
, header
, hdrlen
);
376 datalen
= write_large_blob_data(st
, f
, &entry
->idx
.oid
);
379 hashwrite(f
, buf
, datalen
);
383 return hdrlen
+ datalen
;
386 /* Return 0 if we will bust the pack-size limit */
387 static off_t
write_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
388 unsigned long limit
, int usable_delta
)
390 struct packed_git
*p
= IN_PACK(entry
);
391 struct pack_window
*w_curs
= NULL
;
392 struct revindex_entry
*revidx
;
394 enum object_type type
= oe_type(entry
);
396 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
397 dheader
[MAX_PACK_OBJECT_HEADER
];
399 const unsigned hashsz
= the_hash_algo
->rawsz
;
400 unsigned long entry_size
= SIZE(entry
);
403 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
404 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
405 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
408 offset
= entry
->in_pack_offset
;
409 revidx
= find_pack_revindex(p
, offset
);
410 datalen
= revidx
[1].offset
- offset
;
411 if (!pack_to_stdout
&& p
->index_version
> 1 &&
412 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
)) {
413 error("bad packed object CRC for %s",
414 oid_to_hex(&entry
->idx
.oid
));
416 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
419 offset
+= entry
->in_pack_header_size
;
420 datalen
-= entry
->in_pack_header_size
;
422 if (!pack_to_stdout
&& p
->index_version
== 1 &&
423 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry_size
)) {
424 error("corrupt packed object for %s",
425 oid_to_hex(&entry
->idx
.oid
));
427 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
430 if (type
== OBJ_OFS_DELTA
) {
431 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
432 unsigned pos
= sizeof(dheader
) - 1;
433 dheader
[pos
] = ofs
& 127;
435 dheader
[--pos
] = 128 | (--ofs
& 127);
436 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
440 hashwrite(f
, header
, hdrlen
);
441 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
442 hdrlen
+= sizeof(dheader
) - pos
;
444 } else if (type
== OBJ_REF_DELTA
) {
445 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
449 hashwrite(f
, header
, hdrlen
);
450 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
454 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
458 hashwrite(f
, header
, hdrlen
);
460 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
463 return hdrlen
+ datalen
;
466 /* Return 0 if we will bust the pack-size limit */
467 static off_t
write_object(struct hashfile
*f
,
468 struct object_entry
*entry
,
473 int usable_delta
, to_reuse
;
478 /* apply size limit if limited packsize and not first object */
479 if (!pack_size_limit
|| !nr_written
)
481 else if (pack_size_limit
<= write_offset
)
483 * the earlier object did not fit the limit; avoid
484 * mistaking this with unlimited (i.e. limit = 0).
488 limit
= pack_size_limit
- write_offset
;
491 usable_delta
= 0; /* no delta */
492 else if (!pack_size_limit
)
493 usable_delta
= 1; /* unlimited packfile */
494 else if (DELTA(entry
)->idx
.offset
== (off_t
)-1)
495 usable_delta
= 0; /* base was written to another pack */
496 else if (DELTA(entry
)->idx
.offset
)
497 usable_delta
= 1; /* base already exists in this pack */
499 usable_delta
= 0; /* base could end up in another pack */
502 to_reuse
= 0; /* explicit */
503 else if (!IN_PACK(entry
))
504 to_reuse
= 0; /* can't reuse what we don't have */
505 else if (oe_type(entry
) == OBJ_REF_DELTA
||
506 oe_type(entry
) == OBJ_OFS_DELTA
)
507 /* check_object() decided it for us ... */
508 to_reuse
= usable_delta
;
509 /* ... but pack split may override that */
510 else if (oe_type(entry
) != entry
->in_pack_type
)
511 to_reuse
= 0; /* pack has delta which is unusable */
512 else if (DELTA(entry
))
513 to_reuse
= 0; /* we want to pack afresh */
515 to_reuse
= 1; /* we have it in-pack undeltified,
516 * and we do not need to deltify it.
520 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
522 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
530 entry
->idx
.crc32
= crc32_end(f
);
534 enum write_one_status
{
535 WRITE_ONE_SKIP
= -1, /* already written */
536 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
537 WRITE_ONE_WRITTEN
= 1, /* normal */
538 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
541 static enum write_one_status
write_one(struct hashfile
*f
,
542 struct object_entry
*e
,
549 * we set offset to 1 (which is an impossible value) to mark
550 * the fact that this object is involved in "write its base
551 * first before writing a deltified object" recursion.
553 recursing
= (e
->idx
.offset
== 1);
555 warning("recursive delta detected for object %s",
556 oid_to_hex(&e
->idx
.oid
));
557 return WRITE_ONE_RECURSIVE
;
558 } else if (e
->idx
.offset
|| e
->preferred_base
) {
559 /* offset is non zero if object is written already. */
560 return WRITE_ONE_SKIP
;
563 /* if we are deltified, write out base object first. */
565 e
->idx
.offset
= 1; /* now recurse */
566 switch (write_one(f
, DELTA(e
), offset
)) {
567 case WRITE_ONE_RECURSIVE
:
568 /* we cannot depend on this one */
573 case WRITE_ONE_BREAK
:
574 e
->idx
.offset
= recursing
;
575 return WRITE_ONE_BREAK
;
579 e
->idx
.offset
= *offset
;
580 size
= write_object(f
, e
, *offset
);
582 e
->idx
.offset
= recursing
;
583 return WRITE_ONE_BREAK
;
585 written_list
[nr_written
++] = &e
->idx
;
587 /* make sure off_t is sufficiently large not to wrap */
588 if (signed_add_overflows(*offset
, size
))
589 die("pack too large for current definition of off_t");
591 return WRITE_ONE_WRITTEN
;
594 static int mark_tagged(const char *path
, const struct object_id
*oid
, int flag
,
597 struct object_id peeled
;
598 struct object_entry
*entry
= packlist_find(&to_pack
, oid
->hash
, NULL
);
602 if (!peel_ref(path
, &peeled
)) {
603 entry
= packlist_find(&to_pack
, peeled
.hash
, NULL
);
610 static inline void add_to_write_order(struct object_entry
**wo
,
612 struct object_entry
*e
)
614 if (e
->filled
|| oe_layer(&to_pack
, e
) != write_layer
)
620 static void add_descendants_to_write_order(struct object_entry
**wo
,
622 struct object_entry
*e
)
624 int add_to_order
= 1;
627 struct object_entry
*s
;
628 /* add this node... */
629 add_to_write_order(wo
, endp
, e
);
630 /* all its siblings... */
631 for (s
= DELTA_SIBLING(e
); s
; s
= DELTA_SIBLING(s
)) {
632 add_to_write_order(wo
, endp
, s
);
635 /* drop down a level to add left subtree nodes if possible */
636 if (DELTA_CHILD(e
)) {
641 /* our sibling might have some children, it is next */
642 if (DELTA_SIBLING(e
)) {
643 e
= DELTA_SIBLING(e
);
646 /* go back to our parent node */
648 while (e
&& !DELTA_SIBLING(e
)) {
649 /* we're on the right side of a subtree, keep
650 * going up until we can go right again */
654 /* done- we hit our original root node */
657 /* pass it off to sibling at this level */
658 e
= DELTA_SIBLING(e
);
663 static void add_family_to_write_order(struct object_entry
**wo
,
665 struct object_entry
*e
)
667 struct object_entry
*root
;
669 for (root
= e
; DELTA(root
); root
= DELTA(root
))
671 add_descendants_to_write_order(wo
, endp
, root
);
674 static void compute_layer_order(struct object_entry
**wo
, unsigned int *wo_end
)
676 unsigned int i
, last_untagged
;
677 struct object_entry
*objects
= to_pack
.objects
;
679 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
680 if (objects
[i
].tagged
)
682 add_to_write_order(wo
, wo_end
, &objects
[i
]);
687 * Then fill all the tagged tips.
689 for (; i
< to_pack
.nr_objects
; i
++) {
690 if (objects
[i
].tagged
)
691 add_to_write_order(wo
, wo_end
, &objects
[i
]);
695 * And then all remaining commits and tags.
697 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
698 if (oe_type(&objects
[i
]) != OBJ_COMMIT
&&
699 oe_type(&objects
[i
]) != OBJ_TAG
)
701 add_to_write_order(wo
, wo_end
, &objects
[i
]);
705 * And then all the trees.
707 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
708 if (oe_type(&objects
[i
]) != OBJ_TREE
)
710 add_to_write_order(wo
, wo_end
, &objects
[i
]);
714 * Finally all the rest in really tight order
716 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
717 if (!objects
[i
].filled
&& oe_layer(&to_pack
, &objects
[i
]) == write_layer
)
718 add_family_to_write_order(wo
, wo_end
, &objects
[i
]);
722 static struct object_entry
**compute_write_order(void)
724 uint32_t max_layers
= 1;
725 unsigned int i
, wo_end
;
727 struct object_entry
**wo
;
728 struct object_entry
*objects
= to_pack
.objects
;
730 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
731 objects
[i
].tagged
= 0;
732 objects
[i
].filled
= 0;
733 SET_DELTA_CHILD(&objects
[i
], NULL
);
734 SET_DELTA_SIBLING(&objects
[i
], NULL
);
738 * Fully connect delta_child/delta_sibling network.
739 * Make sure delta_sibling is sorted in the original
742 for (i
= to_pack
.nr_objects
; i
> 0;) {
743 struct object_entry
*e
= &objects
[--i
];
746 /* Mark me as the first child */
747 e
->delta_sibling_idx
= DELTA(e
)->delta_child_idx
;
748 SET_DELTA_CHILD(DELTA(e
), e
);
752 * Mark objects that are at the tip of tags.
754 for_each_tag_ref(mark_tagged
, NULL
);
756 if (use_delta_islands
)
757 max_layers
= compute_pack_layers(&to_pack
);
759 ALLOC_ARRAY(wo
, to_pack
.nr_objects
);
762 for (; write_layer
< max_layers
; ++write_layer
)
763 compute_layer_order(wo
, &wo_end
);
765 if (wo_end
!= to_pack
.nr_objects
)
766 die("ordered %u objects, expected %"PRIu32
, wo_end
, to_pack
.nr_objects
);
771 static off_t
write_reused_pack(struct hashfile
*f
)
773 unsigned char buffer
[8192];
774 off_t to_write
, total
;
777 if (!is_pack_valid(reuse_packfile
))
778 die("packfile is invalid: %s", reuse_packfile
->pack_name
);
780 fd
= git_open(reuse_packfile
->pack_name
);
782 die_errno("unable to open packfile for reuse: %s",
783 reuse_packfile
->pack_name
);
785 if (lseek(fd
, sizeof(struct pack_header
), SEEK_SET
) == -1)
786 die_errno("unable to seek in reused packfile");
788 if (reuse_packfile_offset
< 0)
789 reuse_packfile_offset
= reuse_packfile
->pack_size
- the_hash_algo
->rawsz
;
791 total
= to_write
= reuse_packfile_offset
- sizeof(struct pack_header
);
794 int read_pack
= xread(fd
, buffer
, sizeof(buffer
));
797 die_errno("unable to read from reused packfile");
799 if (read_pack
> to_write
)
800 read_pack
= to_write
;
802 hashwrite(f
, buffer
, read_pack
);
803 to_write
-= read_pack
;
806 * We don't know the actual number of objects written,
807 * only how many bytes written, how many bytes total, and
808 * how many objects total. So we can fake it by pretending all
809 * objects we are writing are the same size. This gives us a
810 * smooth progress meter, and at the end it matches the true
813 written
= reuse_packfile_objects
*
814 (((double)(total
- to_write
)) / total
);
815 display_progress(progress_state
, written
);
819 written
= reuse_packfile_objects
;
820 display_progress(progress_state
, written
);
821 return reuse_packfile_offset
- sizeof(struct pack_header
);
824 static const char no_split_warning
[] = N_(
825 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
828 static void write_pack_file(void)
833 uint32_t nr_remaining
= nr_result
;
834 time_t last_mtime
= 0;
835 struct object_entry
**write_order
;
837 if (progress
> pack_to_stdout
)
838 progress_state
= start_progress(_("Writing objects"), nr_result
);
839 ALLOC_ARRAY(written_list
, to_pack
.nr_objects
);
840 write_order
= compute_write_order();
843 struct object_id oid
;
844 char *pack_tmp_name
= NULL
;
847 f
= hashfd_throughput(1, "<stdout>", progress_state
);
849 f
= create_tmp_packfile(&pack_tmp_name
);
851 offset
= write_pack_header(f
, nr_remaining
);
853 if (reuse_packfile
) {
855 assert(pack_to_stdout
);
857 packfile_size
= write_reused_pack(f
);
858 offset
+= packfile_size
;
862 for (; i
< to_pack
.nr_objects
; i
++) {
863 struct object_entry
*e
= write_order
[i
];
864 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
866 display_progress(progress_state
, written
);
870 * Did we write the wrong # entries in the header?
871 * If so, rewrite it like in fast-import
873 if (pack_to_stdout
) {
874 finalize_hashfile(f
, oid
.hash
, CSUM_HASH_IN_STREAM
| CSUM_CLOSE
);
875 } else if (nr_written
== nr_remaining
) {
876 finalize_hashfile(f
, oid
.hash
, CSUM_HASH_IN_STREAM
| CSUM_FSYNC
| CSUM_CLOSE
);
878 int fd
= finalize_hashfile(f
, oid
.hash
, 0);
879 fixup_pack_header_footer(fd
, oid
.hash
, pack_tmp_name
,
880 nr_written
, oid
.hash
, offset
);
882 if (write_bitmap_index
) {
883 warning(_(no_split_warning
));
884 write_bitmap_index
= 0;
888 if (!pack_to_stdout
) {
890 struct strbuf tmpname
= STRBUF_INIT
;
893 * Packs are runtime accessed in their mtime
894 * order since newer packs are more likely to contain
895 * younger objects. So if we are creating multiple
896 * packs then we should modify the mtime of later ones
897 * to preserve this property.
899 if (stat(pack_tmp_name
, &st
) < 0) {
900 warning_errno("failed to stat %s", pack_tmp_name
);
901 } else if (!last_mtime
) {
902 last_mtime
= st
.st_mtime
;
905 utb
.actime
= st
.st_atime
;
906 utb
.modtime
= --last_mtime
;
907 if (utime(pack_tmp_name
, &utb
) < 0)
908 warning_errno("failed utime() on %s", pack_tmp_name
);
911 strbuf_addf(&tmpname
, "%s-", base_name
);
913 if (write_bitmap_index
) {
914 bitmap_writer_set_checksum(oid
.hash
);
915 bitmap_writer_build_type_index(
916 &to_pack
, written_list
, nr_written
);
919 finish_tmp_packfile(&tmpname
, pack_tmp_name
,
920 written_list
, nr_written
,
921 &pack_idx_opts
, oid
.hash
);
923 if (write_bitmap_index
) {
924 strbuf_addf(&tmpname
, "%s.bitmap", oid_to_hex(&oid
));
926 stop_progress(&progress_state
);
928 bitmap_writer_show_progress(progress
);
929 bitmap_writer_reuse_bitmaps(&to_pack
);
930 bitmap_writer_select_commits(indexed_commits
, indexed_commits_nr
, -1);
931 bitmap_writer_build(&to_pack
);
932 bitmap_writer_finish(written_list
, nr_written
,
933 tmpname
.buf
, write_bitmap_options
);
934 write_bitmap_index
= 0;
937 strbuf_release(&tmpname
);
939 puts(oid_to_hex(&oid
));
942 /* mark written objects as written to previous pack */
943 for (j
= 0; j
< nr_written
; j
++) {
944 written_list
[j
]->offset
= (off_t
)-1;
946 nr_remaining
-= nr_written
;
947 } while (nr_remaining
&& i
< to_pack
.nr_objects
);
951 stop_progress(&progress_state
);
952 if (written
!= nr_result
)
953 die("wrote %"PRIu32
" objects while expecting %"PRIu32
,
957 static int no_try_delta(const char *path
)
959 static struct attr_check
*check
;
962 check
= attr_check_initl("delta", NULL
);
963 if (git_check_attr(path
, check
))
965 if (ATTR_FALSE(check
->items
[0].value
))
971 * When adding an object, check whether we have already added it
972 * to our packing list. If so, we can skip. However, if we are
973 * being asked to excludei t, but the previous mention was to include
974 * it, make sure to adjust its flags and tweak our numbers accordingly.
976 * As an optimization, we pass out the index position where we would have
977 * found the item, since that saves us from having to look it up again a
978 * few lines later when we want to add the new entry.
980 static int have_duplicate_entry(const struct object_id
*oid
,
984 struct object_entry
*entry
;
986 entry
= packlist_find(&to_pack
, oid
->hash
, index_pos
);
991 if (!entry
->preferred_base
)
993 entry
->preferred_base
= 1;
999 static int want_found_object(int exclude
, struct packed_git
*p
)
1007 * When asked to do --local (do not include an object that appears in a
1008 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1009 * an object that appears in a pack marked with .keep), finding a pack
1010 * that matches the criteria is sufficient for us to decide to omit it.
1011 * However, even if this pack does not satisfy the criteria, we need to
1012 * make sure no copy of this object appears in _any_ pack that makes us
1013 * to omit the object, so we need to check all the packs.
1015 * We can however first check whether these options can possible matter;
1016 * if they do not matter we know we want the object in generated pack.
1017 * Otherwise, we signal "-1" at the end to tell the caller that we do
1018 * not know either way, and it needs to check more packs.
1020 if (!ignore_packed_keep_on_disk
&&
1021 !ignore_packed_keep_in_core
&&
1022 (!local
|| !have_non_local_packs
))
1025 if (local
&& !p
->pack_local
)
1027 if (p
->pack_local
&&
1028 ((ignore_packed_keep_on_disk
&& p
->pack_keep
) ||
1029 (ignore_packed_keep_in_core
&& p
->pack_keep_in_core
)))
1032 /* we don't know yet; keep looking for more packs */
1037 * Check whether we want the object in the pack (e.g., we do not want
1038 * objects found in non-local stores if the "--local" option was used).
1040 * If the caller already knows an existing pack it wants to take the object
1041 * from, that is passed in *found_pack and *found_offset; otherwise this
1042 * function finds if there is any pack that has the object and returns the pack
1043 * and its offset in these variables.
1045 static int want_object_in_pack(const struct object_id
*oid
,
1047 struct packed_git
**found_pack
,
1048 off_t
*found_offset
)
1051 struct list_head
*pos
;
1053 if (!exclude
&& local
&& has_loose_object_nonlocal(oid
))
1057 * If we already know the pack object lives in, start checks from that
1058 * pack - in the usual case when neither --local was given nor .keep files
1059 * are present we will determine the answer right now.
1062 want
= want_found_object(exclude
, *found_pack
);
1066 list_for_each(pos
, get_packed_git_mru(the_repository
)) {
1067 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1070 if (p
== *found_pack
)
1071 offset
= *found_offset
;
1073 offset
= find_pack_entry_one(oid
->hash
, p
);
1077 if (!is_pack_valid(p
))
1079 *found_offset
= offset
;
1082 want
= want_found_object(exclude
, p
);
1083 if (!exclude
&& want
> 0)
1085 get_packed_git_mru(the_repository
));
1094 static void create_object_entry(const struct object_id
*oid
,
1095 enum object_type type
,
1100 struct packed_git
*found_pack
,
1103 struct object_entry
*entry
;
1105 entry
= packlist_alloc(&to_pack
, oid
->hash
, index_pos
);
1107 oe_set_type(entry
, type
);
1109 entry
->preferred_base
= 1;
1113 oe_set_in_pack(&to_pack
, entry
, found_pack
);
1114 entry
->in_pack_offset
= found_offset
;
1117 entry
->no_try_delta
= no_try_delta
;
1120 static const char no_closure_warning
[] = N_(
1121 "disabling bitmap writing, as some objects are not being packed"
1124 static int add_object_entry(const struct object_id
*oid
, enum object_type type
,
1125 const char *name
, int exclude
)
1127 struct packed_git
*found_pack
= NULL
;
1128 off_t found_offset
= 0;
1131 display_progress(progress_state
, ++nr_seen
);
1133 if (have_duplicate_entry(oid
, exclude
, &index_pos
))
1136 if (!want_object_in_pack(oid
, exclude
, &found_pack
, &found_offset
)) {
1137 /* The pack is missing an object, so it will not have closure */
1138 if (write_bitmap_index
) {
1139 warning(_(no_closure_warning
));
1140 write_bitmap_index
= 0;
1145 create_object_entry(oid
, type
, pack_name_hash(name
),
1146 exclude
, name
&& no_try_delta(name
),
1147 index_pos
, found_pack
, found_offset
);
1151 static int add_object_entry_from_bitmap(const struct object_id
*oid
,
1152 enum object_type type
,
1153 int flags
, uint32_t name_hash
,
1154 struct packed_git
*pack
, off_t offset
)
1158 display_progress(progress_state
, ++nr_seen
);
1160 if (have_duplicate_entry(oid
, 0, &index_pos
))
1163 if (!want_object_in_pack(oid
, 0, &pack
, &offset
))
1166 create_object_entry(oid
, type
, name_hash
, 0, 0, index_pos
, pack
, offset
);
1170 struct pbase_tree_cache
{
1171 struct object_id oid
;
1175 unsigned long tree_size
;
1178 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1179 static int pbase_tree_cache_ix(const struct object_id
*oid
)
1181 return oid
->hash
[0] % ARRAY_SIZE(pbase_tree_cache
);
1183 static int pbase_tree_cache_ix_incr(int ix
)
1185 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1188 static struct pbase_tree
{
1189 struct pbase_tree
*next
;
1190 /* This is a phony "cache" entry; we are not
1191 * going to evict it or find it through _get()
1192 * mechanism -- this is for the toplevel node that
1193 * would almost always change with any commit.
1195 struct pbase_tree_cache pcache
;
1198 static struct pbase_tree_cache
*pbase_tree_get(const struct object_id
*oid
)
1200 struct pbase_tree_cache
*ent
, *nent
;
1203 enum object_type type
;
1205 int my_ix
= pbase_tree_cache_ix(oid
);
1206 int available_ix
= -1;
1208 /* pbase-tree-cache acts as a limited hashtable.
1209 * your object will be found at your index or within a few
1210 * slots after that slot if it is cached.
1212 for (neigh
= 0; neigh
< 8; neigh
++) {
1213 ent
= pbase_tree_cache
[my_ix
];
1214 if (ent
&& !oidcmp(&ent
->oid
, oid
)) {
1218 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1219 ((0 <= available_ix
) &&
1220 (!ent
&& pbase_tree_cache
[available_ix
])))
1221 available_ix
= my_ix
;
1224 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1227 /* Did not find one. Either we got a bogus request or
1228 * we need to read and perhaps cache.
1230 data
= read_object_file(oid
, &type
, &size
);
1233 if (type
!= OBJ_TREE
) {
1238 /* We need to either cache or return a throwaway copy */
1240 if (available_ix
< 0)
1243 ent
= pbase_tree_cache
[available_ix
];
1244 my_ix
= available_ix
;
1248 nent
= xmalloc(sizeof(*nent
));
1249 nent
->temporary
= (available_ix
< 0);
1252 /* evict and reuse */
1253 free(ent
->tree_data
);
1256 oidcpy(&nent
->oid
, oid
);
1257 nent
->tree_data
= data
;
1258 nent
->tree_size
= size
;
1260 if (!nent
->temporary
)
1261 pbase_tree_cache
[my_ix
] = nent
;
1265 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1267 if (!cache
->temporary
) {
1271 free(cache
->tree_data
);
1275 static int name_cmp_len(const char *name
)
1278 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
1283 static void add_pbase_object(struct tree_desc
*tree
,
1286 const char *fullname
)
1288 struct name_entry entry
;
1291 while (tree_entry(tree
,&entry
)) {
1292 if (S_ISGITLINK(entry
.mode
))
1294 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1295 memcmp(name
, entry
.path
, cmplen
);
1300 if (name
[cmplen
] != '/') {
1301 add_object_entry(entry
.oid
,
1302 object_type(entry
.mode
),
1306 if (S_ISDIR(entry
.mode
)) {
1307 struct tree_desc sub
;
1308 struct pbase_tree_cache
*tree
;
1309 const char *down
= name
+cmplen
+1;
1310 int downlen
= name_cmp_len(down
);
1312 tree
= pbase_tree_get(entry
.oid
);
1315 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1317 add_pbase_object(&sub
, down
, downlen
, fullname
);
1318 pbase_tree_put(tree
);
1323 static unsigned *done_pbase_paths
;
1324 static int done_pbase_paths_num
;
1325 static int done_pbase_paths_alloc
;
1326 static int done_pbase_path_pos(unsigned hash
)
1329 int hi
= done_pbase_paths_num
;
1331 int mi
= lo
+ (hi
- lo
) / 2;
1332 if (done_pbase_paths
[mi
] == hash
)
1334 if (done_pbase_paths
[mi
] < hash
)
1342 static int check_pbase_path(unsigned hash
)
1344 int pos
= done_pbase_path_pos(hash
);
1348 ALLOC_GROW(done_pbase_paths
,
1349 done_pbase_paths_num
+ 1,
1350 done_pbase_paths_alloc
);
1351 done_pbase_paths_num
++;
1352 if (pos
< done_pbase_paths_num
)
1353 MOVE_ARRAY(done_pbase_paths
+ pos
+ 1, done_pbase_paths
+ pos
,
1354 done_pbase_paths_num
- pos
- 1);
1355 done_pbase_paths
[pos
] = hash
;
1359 static void add_preferred_base_object(const char *name
)
1361 struct pbase_tree
*it
;
1363 unsigned hash
= pack_name_hash(name
);
1365 if (!num_preferred_base
|| check_pbase_path(hash
))
1368 cmplen
= name_cmp_len(name
);
1369 for (it
= pbase_tree
; it
; it
= it
->next
) {
1371 add_object_entry(&it
->pcache
.oid
, OBJ_TREE
, NULL
, 1);
1374 struct tree_desc tree
;
1375 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1376 add_pbase_object(&tree
, name
, cmplen
, name
);
1381 static void add_preferred_base(struct object_id
*oid
)
1383 struct pbase_tree
*it
;
1386 struct object_id tree_oid
;
1388 if (window
<= num_preferred_base
++)
1391 data
= read_object_with_reference(oid
, tree_type
, &size
, &tree_oid
);
1395 for (it
= pbase_tree
; it
; it
= it
->next
) {
1396 if (!oidcmp(&it
->pcache
.oid
, &tree_oid
)) {
1402 it
= xcalloc(1, sizeof(*it
));
1403 it
->next
= pbase_tree
;
1406 oidcpy(&it
->pcache
.oid
, &tree_oid
);
1407 it
->pcache
.tree_data
= data
;
1408 it
->pcache
.tree_size
= size
;
1411 static void cleanup_preferred_base(void)
1413 struct pbase_tree
*it
;
1419 struct pbase_tree
*tmp
= it
;
1421 free(tmp
->pcache
.tree_data
);
1425 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1426 if (!pbase_tree_cache
[i
])
1428 free(pbase_tree_cache
[i
]->tree_data
);
1429 FREE_AND_NULL(pbase_tree_cache
[i
]);
1432 FREE_AND_NULL(done_pbase_paths
);
1433 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1436 static void check_object(struct object_entry
*entry
)
1438 unsigned long canonical_size
;
1440 if (IN_PACK(entry
)) {
1441 struct packed_git
*p
= IN_PACK(entry
);
1442 struct pack_window
*w_curs
= NULL
;
1443 const unsigned char *base_ref
= NULL
;
1444 struct object_entry
*base_entry
;
1445 unsigned long used
, used_0
;
1446 unsigned long avail
;
1448 unsigned char *buf
, c
;
1449 enum object_type type
;
1450 unsigned long in_pack_size
;
1452 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1455 * We want in_pack_type even if we do not reuse delta
1456 * since non-delta representations could still be reused.
1458 used
= unpack_object_header_buffer(buf
, avail
,
1465 BUG("invalid type %d", type
);
1466 entry
->in_pack_type
= type
;
1469 * Determine if this is a delta and if so whether we can
1470 * reuse it or not. Otherwise let's find out as cheaply as
1471 * possible what the actual type and size for this object is.
1473 switch (entry
->in_pack_type
) {
1475 /* Not a delta hence we've already got all we need. */
1476 oe_set_type(entry
, entry
->in_pack_type
);
1477 SET_SIZE(entry
, in_pack_size
);
1478 entry
->in_pack_header_size
= used
;
1479 if (oe_type(entry
) < OBJ_COMMIT
|| oe_type(entry
) > OBJ_BLOB
)
1481 unuse_pack(&w_curs
);
1484 if (reuse_delta
&& !entry
->preferred_base
)
1485 base_ref
= use_pack(p
, &w_curs
,
1486 entry
->in_pack_offset
+ used
, NULL
);
1487 entry
->in_pack_header_size
= used
+ the_hash_algo
->rawsz
;
1490 buf
= use_pack(p
, &w_curs
,
1491 entry
->in_pack_offset
+ used
, NULL
);
1497 if (!ofs
|| MSB(ofs
, 7)) {
1498 error("delta base offset overflow in pack for %s",
1499 oid_to_hex(&entry
->idx
.oid
));
1503 ofs
= (ofs
<< 7) + (c
& 127);
1505 ofs
= entry
->in_pack_offset
- ofs
;
1506 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1507 error("delta base offset out of bound for %s",
1508 oid_to_hex(&entry
->idx
.oid
));
1511 if (reuse_delta
&& !entry
->preferred_base
) {
1512 struct revindex_entry
*revidx
;
1513 revidx
= find_pack_revindex(p
, ofs
);
1516 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1518 entry
->in_pack_header_size
= used
+ used_0
;
1522 if (base_ref
&& (base_entry
= packlist_find(&to_pack
, base_ref
, NULL
)) &&
1523 in_same_island(&entry
->idx
.oid
, &base_entry
->idx
.oid
)) {
1525 * If base_ref was set above that means we wish to
1526 * reuse delta data, and we even found that base
1527 * in the list of objects we want to pack. Goodie!
1529 * Depth value does not matter - find_deltas() will
1530 * never consider reused delta as the base object to
1531 * deltify other objects against, in order to avoid
1534 oe_set_type(entry
, entry
->in_pack_type
);
1535 SET_SIZE(entry
, in_pack_size
); /* delta size */
1536 SET_DELTA(entry
, base_entry
);
1537 SET_DELTA_SIZE(entry
, in_pack_size
);
1538 entry
->delta_sibling_idx
= base_entry
->delta_child_idx
;
1539 SET_DELTA_CHILD(base_entry
, entry
);
1540 unuse_pack(&w_curs
);
1544 if (oe_type(entry
)) {
1548 * This must be a delta and we already know what the
1549 * final object type is. Let's extract the actual
1550 * object size from the delta header.
1552 delta_pos
= entry
->in_pack_offset
+ entry
->in_pack_header_size
;
1553 canonical_size
= get_size_from_delta(p
, &w_curs
, delta_pos
);
1554 if (canonical_size
== 0)
1556 SET_SIZE(entry
, canonical_size
);
1557 unuse_pack(&w_curs
);
1562 * No choice but to fall back to the recursive delta walk
1563 * with sha1_object_info() to find about the object type
1567 unuse_pack(&w_curs
);
1571 oid_object_info(the_repository
, &entry
->idx
.oid
, &canonical_size
));
1572 if (entry
->type_valid
) {
1573 SET_SIZE(entry
, canonical_size
);
1576 * Bad object type is checked in prepare_pack(). This is
1577 * to permit a missing preferred base object to be ignored
1578 * as a preferred base. Doing so can result in a larger
1579 * pack file, but the transfer will still take place.
1584 static int pack_offset_sort(const void *_a
, const void *_b
)
1586 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1587 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1588 const struct packed_git
*a_in_pack
= IN_PACK(a
);
1589 const struct packed_git
*b_in_pack
= IN_PACK(b
);
1591 /* avoid filesystem trashing with loose objects */
1592 if (!a_in_pack
&& !b_in_pack
)
1593 return oidcmp(&a
->idx
.oid
, &b
->idx
.oid
);
1595 if (a_in_pack
< b_in_pack
)
1597 if (a_in_pack
> b_in_pack
)
1599 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1600 (a
->in_pack_offset
> b
->in_pack_offset
);
1604 * Drop an on-disk delta we were planning to reuse. Naively, this would
1605 * just involve blanking out the "delta" field, but we have to deal
1606 * with some extra book-keeping:
1608 * 1. Removing ourselves from the delta_sibling linked list.
1610 * 2. Updating our size/type to the non-delta representation. These were
1611 * either not recorded initially (size) or overwritten with the delta type
1612 * (type) when check_object() decided to reuse the delta.
1614 * 3. Resetting our delta depth, as we are now a base object.
1616 static void drop_reused_delta(struct object_entry
*entry
)
1618 unsigned *idx
= &to_pack
.objects
[entry
->delta_idx
- 1].delta_child_idx
;
1619 struct object_info oi
= OBJECT_INFO_INIT
;
1620 enum object_type type
;
1624 struct object_entry
*oe
= &to_pack
.objects
[*idx
- 1];
1627 *idx
= oe
->delta_sibling_idx
;
1629 idx
= &oe
->delta_sibling_idx
;
1631 SET_DELTA(entry
, NULL
);
1636 if (packed_object_info(the_repository
, IN_PACK(entry
), entry
->in_pack_offset
, &oi
) < 0) {
1638 * We failed to get the info from this pack for some reason;
1639 * fall back to sha1_object_info, which may find another copy.
1640 * And if that fails, the error will be recorded in oe_type(entry)
1641 * and dealt with in prepare_pack().
1644 oid_object_info(the_repository
, &entry
->idx
.oid
, &size
));
1646 oe_set_type(entry
, type
);
1648 SET_SIZE(entry
, size
);
1652 * Follow the chain of deltas from this entry onward, throwing away any links
1653 * that cause us to hit a cycle (as determined by the DFS state flags in
1656 * We also detect too-long reused chains that would violate our --depth
1659 static void break_delta_chains(struct object_entry
*entry
)
1662 * The actual depth of each object we will write is stored as an int,
1663 * as it cannot exceed our int "depth" limit. But before we break
1664 * changes based no that limit, we may potentially go as deep as the
1665 * number of objects, which is elsewhere bounded to a uint32_t.
1667 uint32_t total_depth
;
1668 struct object_entry
*cur
, *next
;
1670 for (cur
= entry
, total_depth
= 0;
1672 cur
= DELTA(cur
), total_depth
++) {
1673 if (cur
->dfs_state
== DFS_DONE
) {
1675 * We've already seen this object and know it isn't
1676 * part of a cycle. We do need to append its depth
1679 total_depth
+= cur
->depth
;
1684 * We break cycles before looping, so an ACTIVE state (or any
1685 * other cruft which made its way into the state variable)
1688 if (cur
->dfs_state
!= DFS_NONE
)
1689 BUG("confusing delta dfs state in first pass: %d",
1693 * Now we know this is the first time we've seen the object. If
1694 * it's not a delta, we're done traversing, but we'll mark it
1695 * done to save time on future traversals.
1698 cur
->dfs_state
= DFS_DONE
;
1703 * Mark ourselves as active and see if the next step causes
1704 * us to cycle to another active object. It's important to do
1705 * this _before_ we loop, because it impacts where we make the
1706 * cut, and thus how our total_depth counter works.
1707 * E.g., We may see a partial loop like:
1709 * A -> B -> C -> D -> B
1711 * Cutting B->C breaks the cycle. But now the depth of A is
1712 * only 1, and our total_depth counter is at 3. The size of the
1713 * error is always one less than the size of the cycle we
1714 * broke. Commits C and D were "lost" from A's chain.
1716 * If we instead cut D->B, then the depth of A is correct at 3.
1717 * We keep all commits in the chain that we examined.
1719 cur
->dfs_state
= DFS_ACTIVE
;
1720 if (DELTA(cur
)->dfs_state
== DFS_ACTIVE
) {
1721 drop_reused_delta(cur
);
1722 cur
->dfs_state
= DFS_DONE
;
1728 * And now that we've gone all the way to the bottom of the chain, we
1729 * need to clear the active flags and set the depth fields as
1730 * appropriate. Unlike the loop above, which can quit when it drops a
1731 * delta, we need to keep going to look for more depth cuts. So we need
1732 * an extra "next" pointer to keep going after we reset cur->delta.
1734 for (cur
= entry
; cur
; cur
= next
) {
1738 * We should have a chain of zero or more ACTIVE states down to
1739 * a final DONE. We can quit after the DONE, because either it
1740 * has no bases, or we've already handled them in a previous
1743 if (cur
->dfs_state
== DFS_DONE
)
1745 else if (cur
->dfs_state
!= DFS_ACTIVE
)
1746 BUG("confusing delta dfs state in second pass: %d",
1750 * If the total_depth is more than depth, then we need to snip
1751 * the chain into two or more smaller chains that don't exceed
1752 * the maximum depth. Most of the resulting chains will contain
1753 * (depth + 1) entries (i.e., depth deltas plus one base), and
1754 * the last chain (i.e., the one containing entry) will contain
1755 * whatever entries are left over, namely
1756 * (total_depth % (depth + 1)) of them.
1758 * Since we are iterating towards decreasing depth, we need to
1759 * decrement total_depth as we go, and we need to write to the
1760 * entry what its final depth will be after all of the
1761 * snipping. Since we're snipping into chains of length (depth
1762 * + 1) entries, the final depth of an entry will be its
1763 * original depth modulo (depth + 1). Any time we encounter an
1764 * entry whose final depth is supposed to be zero, we snip it
1765 * from its delta base, thereby making it so.
1767 cur
->depth
= (total_depth
--) % (depth
+ 1);
1769 drop_reused_delta(cur
);
1771 cur
->dfs_state
= DFS_DONE
;
1775 static void get_object_details(void)
1778 struct object_entry
**sorted_by_offset
;
1781 progress_state
= start_progress(_("Counting objects"),
1782 to_pack
.nr_objects
);
1784 sorted_by_offset
= xcalloc(to_pack
.nr_objects
, sizeof(struct object_entry
*));
1785 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1786 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
1787 QSORT(sorted_by_offset
, to_pack
.nr_objects
, pack_offset_sort
);
1789 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
1790 struct object_entry
*entry
= sorted_by_offset
[i
];
1791 check_object(entry
);
1792 if (entry
->type_valid
&&
1793 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
))
1794 entry
->no_try_delta
= 1;
1795 display_progress(progress_state
, i
+ 1);
1797 stop_progress(&progress_state
);
1800 * This must happen in a second pass, since we rely on the delta
1801 * information for the whole list being completed.
1803 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1804 break_delta_chains(&to_pack
.objects
[i
]);
1806 free(sorted_by_offset
);
1810 * We search for deltas in a list sorted by type, by filename hash, and then
1811 * by size, so that we see progressively smaller and smaller files.
1812 * That's because we prefer deltas to be from the bigger file
1813 * to the smaller -- deletes are potentially cheaper, but perhaps
1814 * more importantly, the bigger file is likely the more recent
1815 * one. The deepest deltas are therefore the oldest objects which are
1816 * less susceptible to be accessed often.
1818 static int type_size_sort(const void *_a
, const void *_b
)
1820 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1821 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1822 enum object_type a_type
= oe_type(a
);
1823 enum object_type b_type
= oe_type(b
);
1824 unsigned long a_size
= SIZE(a
);
1825 unsigned long b_size
= SIZE(b
);
1827 if (a_type
> b_type
)
1829 if (a_type
< b_type
)
1831 if (a
->hash
> b
->hash
)
1833 if (a
->hash
< b
->hash
)
1835 if (a
->preferred_base
> b
->preferred_base
)
1837 if (a
->preferred_base
< b
->preferred_base
)
1839 if (use_delta_islands
) {
1840 int island_cmp
= island_delta_cmp(&a
->idx
.oid
, &b
->idx
.oid
);
1844 if (a_size
> b_size
)
1846 if (a_size
< b_size
)
1848 return a
< b
? -1 : (a
> b
); /* newest first */
1852 struct object_entry
*entry
;
1854 struct delta_index
*index
;
1858 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1859 unsigned long delta_size
)
1861 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1864 if (delta_size
< cache_max_small_delta_size
)
1867 /* cache delta, if objects are large enough compared to delta size */
1868 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1876 static pthread_mutex_t read_mutex
;
1877 #define read_lock() pthread_mutex_lock(&read_mutex)
1878 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1880 static pthread_mutex_t cache_mutex
;
1881 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1882 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1884 static pthread_mutex_t progress_mutex
;
1885 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1886 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1890 #define read_lock() (void)0
1891 #define read_unlock() (void)0
1892 #define cache_lock() (void)0
1893 #define cache_unlock() (void)0
1894 #define progress_lock() (void)0
1895 #define progress_unlock() (void)0
1900 * Return the size of the object without doing any delta
1901 * reconstruction (so non-deltas are true object sizes, but deltas
1902 * return the size of the delta data).
1904 unsigned long oe_get_size_slow(struct packing_data
*pack
,
1905 const struct object_entry
*e
)
1907 struct packed_git
*p
;
1908 struct pack_window
*w_curs
;
1910 enum object_type type
;
1911 unsigned long used
, avail
, size
;
1913 if (e
->type_
!= OBJ_OFS_DELTA
&& e
->type_
!= OBJ_REF_DELTA
) {
1915 if (oid_object_info(the_repository
, &e
->idx
.oid
, &size
) < 0)
1916 die(_("unable to get size of %s"),
1917 oid_to_hex(&e
->idx
.oid
));
1922 p
= oe_in_pack(pack
, e
);
1924 BUG("when e->type is a delta, it must belong to a pack");
1928 buf
= use_pack(p
, &w_curs
, e
->in_pack_offset
, &avail
);
1929 used
= unpack_object_header_buffer(buf
, avail
, &type
, &size
);
1931 die(_("unable to parse object header of %s"),
1932 oid_to_hex(&e
->idx
.oid
));
1934 unuse_pack(&w_curs
);
1939 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1940 unsigned max_depth
, unsigned long *mem_usage
)
1942 struct object_entry
*trg_entry
= trg
->entry
;
1943 struct object_entry
*src_entry
= src
->entry
;
1944 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1946 enum object_type type
;
1949 /* Don't bother doing diffs between different types */
1950 if (oe_type(trg_entry
) != oe_type(src_entry
))
1954 * We do not bother to try a delta that we discarded on an
1955 * earlier try, but only when reusing delta data. Note that
1956 * src_entry that is marked as the preferred_base should always
1957 * be considered, as even if we produce a suboptimal delta against
1958 * it, we will still save the transfer cost, as we already know
1959 * the other side has it and we won't send src_entry at all.
1961 if (reuse_delta
&& IN_PACK(trg_entry
) &&
1962 IN_PACK(trg_entry
) == IN_PACK(src_entry
) &&
1963 !src_entry
->preferred_base
&&
1964 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1965 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1968 /* Let's not bust the allowed depth. */
1969 if (src
->depth
>= max_depth
)
1972 /* Now some size filtering heuristics. */
1973 trg_size
= SIZE(trg_entry
);
1974 if (!DELTA(trg_entry
)) {
1975 max_size
= trg_size
/2 - the_hash_algo
->rawsz
;
1978 max_size
= DELTA_SIZE(trg_entry
);
1979 ref_depth
= trg
->depth
;
1981 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
1982 (max_depth
- ref_depth
+ 1);
1985 src_size
= SIZE(src_entry
);
1986 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1987 if (sizediff
>= max_size
)
1989 if (trg_size
< src_size
/ 32)
1992 if (!in_same_island(&trg
->entry
->idx
.oid
, &src
->entry
->idx
.oid
))
1995 /* Load data if not already done */
1998 trg
->data
= read_object_file(&trg_entry
->idx
.oid
, &type
, &sz
);
2001 die("object %s cannot be read",
2002 oid_to_hex(&trg_entry
->idx
.oid
));
2004 die("object %s inconsistent object length (%lu vs %lu)",
2005 oid_to_hex(&trg_entry
->idx
.oid
), sz
,
2011 src
->data
= read_object_file(&src_entry
->idx
.oid
, &type
, &sz
);
2014 if (src_entry
->preferred_base
) {
2015 static int warned
= 0;
2017 warning("object %s cannot be read",
2018 oid_to_hex(&src_entry
->idx
.oid
));
2020 * Those objects are not included in the
2021 * resulting pack. Be resilient and ignore
2022 * them if they can't be read, in case the
2023 * pack could be created nevertheless.
2027 die("object %s cannot be read",
2028 oid_to_hex(&src_entry
->idx
.oid
));
2031 die("object %s inconsistent object length (%lu vs %lu)",
2032 oid_to_hex(&src_entry
->idx
.oid
), sz
,
2037 src
->index
= create_delta_index(src
->data
, src_size
);
2039 static int warned
= 0;
2041 warning("suboptimal pack - out of memory");
2044 *mem_usage
+= sizeof_delta_index(src
->index
);
2047 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
2050 if (delta_size
>= (1U << OE_DELTA_SIZE_BITS
)) {
2055 if (DELTA(trg_entry
)) {
2056 /* Prefer only shallower same-sized deltas. */
2057 if (delta_size
== DELTA_SIZE(trg_entry
) &&
2058 src
->depth
+ 1 >= trg
->depth
) {
2065 * Handle memory allocation outside of the cache
2066 * accounting lock. Compiler will optimize the strangeness
2067 * away when NO_PTHREADS is defined.
2069 free(trg_entry
->delta_data
);
2071 if (trg_entry
->delta_data
) {
2072 delta_cache_size
-= DELTA_SIZE(trg_entry
);
2073 trg_entry
->delta_data
= NULL
;
2075 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
2076 delta_cache_size
+= delta_size
;
2078 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
2084 SET_DELTA(trg_entry
, src_entry
);
2085 SET_DELTA_SIZE(trg_entry
, delta_size
);
2086 trg
->depth
= src
->depth
+ 1;
2091 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
2093 struct object_entry
*child
= DELTA_CHILD(me
);
2096 unsigned int c
= check_delta_limit(child
, n
+ 1);
2099 child
= DELTA_SIBLING(child
);
2104 static unsigned long free_unpacked(struct unpacked
*n
)
2106 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
2107 free_delta_index(n
->index
);
2110 freed_mem
+= SIZE(n
->entry
);
2111 FREE_AND_NULL(n
->data
);
2118 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
2119 int window
, int depth
, unsigned *processed
)
2121 uint32_t i
, idx
= 0, count
= 0;
2122 struct unpacked
*array
;
2123 unsigned long mem_usage
= 0;
2125 array
= xcalloc(window
, sizeof(struct unpacked
));
2128 struct object_entry
*entry
;
2129 struct unpacked
*n
= array
+ idx
;
2130 int j
, max_depth
, best_base
= -1;
2139 if (!entry
->preferred_base
) {
2141 display_progress(progress_state
, *processed
);
2145 mem_usage
-= free_unpacked(n
);
2148 while (window_memory_limit
&&
2149 mem_usage
> window_memory_limit
&&
2151 uint32_t tail
= (idx
+ window
- count
) % window
;
2152 mem_usage
-= free_unpacked(array
+ tail
);
2156 /* We do not compute delta to *create* objects we are not
2159 if (entry
->preferred_base
)
2163 * If the current object is at pack edge, take the depth the
2164 * objects that depend on the current object into account
2165 * otherwise they would become too deep.
2168 if (DELTA_CHILD(entry
)) {
2169 max_depth
-= check_delta_limit(entry
, 0);
2177 uint32_t other_idx
= idx
+ j
;
2179 if (other_idx
>= window
)
2180 other_idx
-= window
;
2181 m
= array
+ other_idx
;
2184 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
2188 best_base
= other_idx
;
2192 * If we decided to cache the delta data, then it is best
2193 * to compress it right away. First because we have to do
2194 * it anyway, and doing it here while we're threaded will
2195 * save a lot of time in the non threaded write phase,
2196 * as well as allow for caching more deltas within
2197 * the same cache size limit.
2199 * But only if not writing to stdout, since in that case
2200 * the network is most likely throttling writes anyway,
2201 * and therefore it is best to go to the write phase ASAP
2202 * instead, as we can afford spending more time compressing
2203 * between writes at that moment.
2205 if (entry
->delta_data
&& !pack_to_stdout
) {
2208 size
= do_compress(&entry
->delta_data
, DELTA_SIZE(entry
));
2209 if (size
< (1U << OE_Z_DELTA_BITS
)) {
2210 entry
->z_delta_size
= size
;
2212 delta_cache_size
-= DELTA_SIZE(entry
);
2213 delta_cache_size
+= entry
->z_delta_size
;
2216 FREE_AND_NULL(entry
->delta_data
);
2217 entry
->z_delta_size
= 0;
2221 /* if we made n a delta, and if n is already at max
2222 * depth, leaving it in the window is pointless. we
2223 * should evict it first.
2225 if (DELTA(entry
) && max_depth
<= n
->depth
)
2229 * Move the best delta base up in the window, after the
2230 * currently deltified object, to keep it longer. It will
2231 * be the first base object to be attempted next.
2234 struct unpacked swap
= array
[best_base
];
2235 int dist
= (window
+ idx
- best_base
) % window
;
2236 int dst
= best_base
;
2238 int src
= (dst
+ 1) % window
;
2239 array
[dst
] = array
[src
];
2247 if (count
+ 1 < window
)
2253 for (i
= 0; i
< window
; ++i
) {
2254 free_delta_index(array
[i
].index
);
2255 free(array
[i
].data
);
2262 static void try_to_free_from_threads(size_t size
)
2265 release_pack_memory(size
);
2269 static try_to_free_t old_try_to_free_routine
;
2272 * The main thread waits on the condition that (at least) one of the workers
2273 * has stopped working (which is indicated in the .working member of
2274 * struct thread_params).
2275 * When a work thread has completed its work, it sets .working to 0 and
2276 * signals the main thread and waits on the condition that .data_ready
2280 struct thread_params
{
2282 struct object_entry
**list
;
2289 pthread_mutex_t mutex
;
2290 pthread_cond_t cond
;
2291 unsigned *processed
;
2294 static pthread_cond_t progress_cond
;
2297 * Mutex and conditional variable can't be statically-initialized on Windows.
2299 static void init_threaded_search(void)
2301 init_recursive_mutex(&read_mutex
);
2302 pthread_mutex_init(&cache_mutex
, NULL
);
2303 pthread_mutex_init(&progress_mutex
, NULL
);
2304 pthread_cond_init(&progress_cond
, NULL
);
2305 old_try_to_free_routine
= set_try_to_free_routine(try_to_free_from_threads
);
2308 static void cleanup_threaded_search(void)
2310 set_try_to_free_routine(old_try_to_free_routine
);
2311 pthread_cond_destroy(&progress_cond
);
2312 pthread_mutex_destroy(&read_mutex
);
2313 pthread_mutex_destroy(&cache_mutex
);
2314 pthread_mutex_destroy(&progress_mutex
);
2317 static void *threaded_find_deltas(void *arg
)
2319 struct thread_params
*me
= arg
;
2322 while (me
->remaining
) {
2325 find_deltas(me
->list
, &me
->remaining
,
2326 me
->window
, me
->depth
, me
->processed
);
2330 pthread_cond_signal(&progress_cond
);
2334 * We must not set ->data_ready before we wait on the
2335 * condition because the main thread may have set it to 1
2336 * before we get here. In order to be sure that new
2337 * work is available if we see 1 in ->data_ready, it
2338 * was initialized to 0 before this thread was spawned
2339 * and we reset it to 0 right away.
2341 pthread_mutex_lock(&me
->mutex
);
2342 while (!me
->data_ready
)
2343 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2345 pthread_mutex_unlock(&me
->mutex
);
2350 /* leave ->working 1 so that this doesn't get more work assigned */
2354 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2355 int window
, int depth
, unsigned *processed
)
2357 struct thread_params
*p
;
2358 int i
, ret
, active_threads
= 0;
2360 init_threaded_search();
2362 if (delta_search_threads
<= 1) {
2363 find_deltas(list
, &list_size
, window
, depth
, processed
);
2364 cleanup_threaded_search();
2367 if (progress
> pack_to_stdout
)
2368 fprintf(stderr
, "Delta compression using up to %d threads.\n",
2369 delta_search_threads
);
2370 p
= xcalloc(delta_search_threads
, sizeof(*p
));
2372 /* Partition the work amongst work threads. */
2373 for (i
= 0; i
< delta_search_threads
; i
++) {
2374 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2376 /* don't use too small segments or no deltas will be found */
2377 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2380 p
[i
].window
= window
;
2382 p
[i
].processed
= processed
;
2384 p
[i
].data_ready
= 0;
2386 /* try to split chunks on "path" boundaries */
2387 while (sub_size
&& sub_size
< list_size
&&
2388 list
[sub_size
]->hash
&&
2389 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2393 p
[i
].list_size
= sub_size
;
2394 p
[i
].remaining
= sub_size
;
2397 list_size
-= sub_size
;
2400 /* Start work threads. */
2401 for (i
= 0; i
< delta_search_threads
; i
++) {
2402 if (!p
[i
].list_size
)
2404 pthread_mutex_init(&p
[i
].mutex
, NULL
);
2405 pthread_cond_init(&p
[i
].cond
, NULL
);
2406 ret
= pthread_create(&p
[i
].thread
, NULL
,
2407 threaded_find_deltas
, &p
[i
]);
2409 die("unable to create thread: %s", strerror(ret
));
2414 * Now let's wait for work completion. Each time a thread is done
2415 * with its work, we steal half of the remaining work from the
2416 * thread with the largest number of unprocessed objects and give
2417 * it to that newly idle thread. This ensure good load balancing
2418 * until the remaining object list segments are simply too short
2419 * to be worth splitting anymore.
2421 while (active_threads
) {
2422 struct thread_params
*target
= NULL
;
2423 struct thread_params
*victim
= NULL
;
2424 unsigned sub_size
= 0;
2428 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
2433 pthread_cond_wait(&progress_cond
, &progress_mutex
);
2436 for (i
= 0; i
< delta_search_threads
; i
++)
2437 if (p
[i
].remaining
> 2*window
&&
2438 (!victim
|| victim
->remaining
< p
[i
].remaining
))
2441 sub_size
= victim
->remaining
/ 2;
2442 list
= victim
->list
+ victim
->list_size
- sub_size
;
2443 while (sub_size
&& list
[0]->hash
&&
2444 list
[0]->hash
== list
[-1]->hash
) {
2450 * It is possible for some "paths" to have
2451 * so many objects that no hash boundary
2452 * might be found. Let's just steal the
2453 * exact half in that case.
2455 sub_size
= victim
->remaining
/ 2;
2458 target
->list
= list
;
2459 victim
->list_size
-= sub_size
;
2460 victim
->remaining
-= sub_size
;
2462 target
->list_size
= sub_size
;
2463 target
->remaining
= sub_size
;
2464 target
->working
= 1;
2467 pthread_mutex_lock(&target
->mutex
);
2468 target
->data_ready
= 1;
2469 pthread_cond_signal(&target
->cond
);
2470 pthread_mutex_unlock(&target
->mutex
);
2473 pthread_join(target
->thread
, NULL
);
2474 pthread_cond_destroy(&target
->cond
);
2475 pthread_mutex_destroy(&target
->mutex
);
2479 cleanup_threaded_search();
2484 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2487 static void add_tag_chain(const struct object_id
*oid
)
2492 * We catch duplicates already in add_object_entry(), but we'd
2493 * prefer to do this extra check to avoid having to parse the
2494 * tag at all if we already know that it's being packed (e.g., if
2495 * it was included via bitmaps, we would not have parsed it
2498 if (packlist_find(&to_pack
, oid
->hash
, NULL
))
2501 tag
= lookup_tag(the_repository
, oid
);
2503 if (!tag
|| parse_tag(tag
) || !tag
->tagged
)
2504 die("unable to pack objects reachable from tag %s",
2507 add_object_entry(&tag
->object
.oid
, OBJ_TAG
, NULL
, 0);
2509 if (tag
->tagged
->type
!= OBJ_TAG
)
2512 tag
= (struct tag
*)tag
->tagged
;
2516 static int add_ref_tag(const char *path
, const struct object_id
*oid
, int flag
, void *cb_data
)
2518 struct object_id peeled
;
2520 if (starts_with(path
, "refs/tags/") && /* is a tag? */
2521 !peel_ref(path
, &peeled
) && /* peelable? */
2522 packlist_find(&to_pack
, peeled
.hash
, NULL
)) /* object packed? */
2527 static void prepare_pack(int window
, int depth
)
2529 struct object_entry
**delta_list
;
2530 uint32_t i
, nr_deltas
;
2533 if (use_delta_islands
)
2534 resolve_tree_islands(progress
, &to_pack
);
2536 get_object_details();
2539 * If we're locally repacking then we need to be doubly careful
2540 * from now on in order to make sure no stealth corruption gets
2541 * propagated to the new pack. Clients receiving streamed packs
2542 * should validate everything they get anyway so no need to incur
2543 * the additional cost here in that case.
2545 if (!pack_to_stdout
)
2546 do_check_packed_object_crc
= 1;
2548 if (!to_pack
.nr_objects
|| !window
|| !depth
)
2551 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
2554 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2555 struct object_entry
*entry
= to_pack
.objects
+ i
;
2558 /* This happens if we decided to reuse existing
2559 * delta from a pack. "reuse_delta &&" is implied.
2563 if (!entry
->type_valid
||
2564 oe_size_less_than(&to_pack
, entry
, 50))
2567 if (entry
->no_try_delta
)
2570 if (!entry
->preferred_base
) {
2572 if (oe_type(entry
) < 0)
2573 die("unable to get type of object %s",
2574 oid_to_hex(&entry
->idx
.oid
));
2576 if (oe_type(entry
) < 0) {
2578 * This object is not found, but we
2579 * don't have to include it anyway.
2585 delta_list
[n
++] = entry
;
2588 if (nr_deltas
&& n
> 1) {
2589 unsigned nr_done
= 0;
2591 progress_state
= start_progress(_("Compressing objects"),
2593 QSORT(delta_list
, n
, type_size_sort
);
2594 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
2595 stop_progress(&progress_state
);
2596 if (nr_done
!= nr_deltas
)
2597 die("inconsistency with delta count");
2602 static int git_pack_config(const char *k
, const char *v
, void *cb
)
2604 if (!strcmp(k
, "pack.window")) {
2605 window
= git_config_int(k
, v
);
2608 if (!strcmp(k
, "pack.windowmemory")) {
2609 window_memory_limit
= git_config_ulong(k
, v
);
2612 if (!strcmp(k
, "pack.depth")) {
2613 depth
= git_config_int(k
, v
);
2616 if (!strcmp(k
, "pack.deltacachesize")) {
2617 max_delta_cache_size
= git_config_int(k
, v
);
2620 if (!strcmp(k
, "pack.deltacachelimit")) {
2621 cache_max_small_delta_size
= git_config_int(k
, v
);
2624 if (!strcmp(k
, "pack.writebitmaphashcache")) {
2625 if (git_config_bool(k
, v
))
2626 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
2628 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
2630 if (!strcmp(k
, "pack.usebitmaps")) {
2631 use_bitmap_index_default
= git_config_bool(k
, v
);
2634 if (!strcmp(k
, "pack.threads")) {
2635 delta_search_threads
= git_config_int(k
, v
);
2636 if (delta_search_threads
< 0)
2637 die("invalid number of threads specified (%d)",
2638 delta_search_threads
);
2640 if (delta_search_threads
!= 1) {
2641 warning("no threads support, ignoring %s", k
);
2642 delta_search_threads
= 0;
2647 if (!strcmp(k
, "pack.indexversion")) {
2648 pack_idx_opts
.version
= git_config_int(k
, v
);
2649 if (pack_idx_opts
.version
> 2)
2650 die("bad pack.indexversion=%"PRIu32
,
2651 pack_idx_opts
.version
);
2654 return git_default_config(k
, v
, cb
);
2657 static void read_object_list_from_stdin(void)
2659 char line
[GIT_MAX_HEXSZ
+ 1 + PATH_MAX
+ 2];
2660 struct object_id oid
;
2664 if (!fgets(line
, sizeof(line
), stdin
)) {
2668 die("fgets returned NULL, not EOF, not error!");
2674 if (line
[0] == '-') {
2675 if (get_oid_hex(line
+1, &oid
))
2676 die("expected edge object ID, got garbage:\n %s",
2678 add_preferred_base(&oid
);
2681 if (parse_oid_hex(line
, &oid
, &p
))
2682 die("expected object ID, got garbage:\n %s", line
);
2684 add_preferred_base_object(p
+ 1);
2685 add_object_entry(&oid
, OBJ_NONE
, p
+ 1, 0);
2689 /* Remember to update object flag allocation in object.h */
2690 #define OBJECT_ADDED (1u<<20)
2692 static void show_commit(struct commit
*commit
, void *data
)
2694 add_object_entry(&commit
->object
.oid
, OBJ_COMMIT
, NULL
, 0);
2695 commit
->object
.flags
|= OBJECT_ADDED
;
2697 if (write_bitmap_index
)
2698 index_commit_for_bitmap(commit
);
2700 if (use_delta_islands
)
2701 propagate_island_marks(commit
);
2704 static void show_object(struct object
*obj
, const char *name
, void *data
)
2706 add_preferred_base_object(name
);
2707 add_object_entry(&obj
->oid
, obj
->type
, name
, 0);
2708 obj
->flags
|= OBJECT_ADDED
;
2710 if (use_delta_islands
) {
2713 struct object_entry
*ent
;
2715 /* the empty string is a root tree, which is depth 0 */
2716 depth
= *name
? 1 : 0;
2717 for (p
= strchr(name
, '/'); p
; p
= strchr(p
+ 1, '/'))
2720 ent
= packlist_find(&to_pack
, obj
->oid
.hash
, NULL
);
2721 if (ent
&& depth
> oe_tree_depth(&to_pack
, ent
))
2722 oe_set_tree_depth(&to_pack
, ent
, depth
);
2726 static void show_object__ma_allow_any(struct object
*obj
, const char *name
, void *data
)
2728 assert(arg_missing_action
== MA_ALLOW_ANY
);
2731 * Quietly ignore ALL missing objects. This avoids problems with
2732 * staging them now and getting an odd error later.
2734 if (!has_object_file(&obj
->oid
))
2737 show_object(obj
, name
, data
);
2740 static void show_object__ma_allow_promisor(struct object
*obj
, const char *name
, void *data
)
2742 assert(arg_missing_action
== MA_ALLOW_PROMISOR
);
2745 * Quietly ignore EXPECTED missing objects. This avoids problems with
2746 * staging them now and getting an odd error later.
2748 if (!has_object_file(&obj
->oid
) && is_promisor_object(&obj
->oid
))
2751 show_object(obj
, name
, data
);
2754 static int option_parse_missing_action(const struct option
*opt
,
2755 const char *arg
, int unset
)
2760 if (!strcmp(arg
, "error")) {
2761 arg_missing_action
= MA_ERROR
;
2762 fn_show_object
= show_object
;
2766 if (!strcmp(arg
, "allow-any")) {
2767 arg_missing_action
= MA_ALLOW_ANY
;
2768 fetch_if_missing
= 0;
2769 fn_show_object
= show_object__ma_allow_any
;
2773 if (!strcmp(arg
, "allow-promisor")) {
2774 arg_missing_action
= MA_ALLOW_PROMISOR
;
2775 fetch_if_missing
= 0;
2776 fn_show_object
= show_object__ma_allow_promisor
;
2780 die(_("invalid value for --missing"));
2784 static void show_edge(struct commit
*commit
)
2786 add_preferred_base(&commit
->object
.oid
);
2789 struct in_pack_object
{
2791 struct object
*object
;
2797 struct in_pack_object
*array
;
2800 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
2802 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->oid
.hash
, p
);
2803 in_pack
->array
[in_pack
->nr
].object
= object
;
2808 * Compare the objects in the offset order, in order to emulate the
2809 * "git rev-list --objects" output that produced the pack originally.
2811 static int ofscmp(const void *a_
, const void *b_
)
2813 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
2814 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
2816 if (a
->offset
< b
->offset
)
2818 else if (a
->offset
> b
->offset
)
2821 return oidcmp(&a
->object
->oid
, &b
->object
->oid
);
2824 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
2826 struct packed_git
*p
;
2827 struct in_pack in_pack
;
2830 memset(&in_pack
, 0, sizeof(in_pack
));
2832 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
2833 struct object_id oid
;
2836 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
2838 if (open_pack_index(p
))
2839 die("cannot open pack index");
2841 ALLOC_GROW(in_pack
.array
,
2842 in_pack
.nr
+ p
->num_objects
,
2845 for (i
= 0; i
< p
->num_objects
; i
++) {
2846 nth_packed_object_oid(&oid
, p
, i
);
2847 o
= lookup_unknown_object(oid
.hash
);
2848 if (!(o
->flags
& OBJECT_ADDED
))
2849 mark_in_pack_object(o
, p
, &in_pack
);
2850 o
->flags
|= OBJECT_ADDED
;
2855 QSORT(in_pack
.array
, in_pack
.nr
, ofscmp
);
2856 for (i
= 0; i
< in_pack
.nr
; i
++) {
2857 struct object
*o
= in_pack
.array
[i
].object
;
2858 add_object_entry(&o
->oid
, o
->type
, "", 0);
2861 free(in_pack
.array
);
2864 static int add_loose_object(const struct object_id
*oid
, const char *path
,
2867 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
2870 warning("loose object at %s could not be examined", path
);
2874 add_object_entry(oid
, type
, "", 0);
2879 * We actually don't even have to worry about reachability here.
2880 * add_object_entry will weed out duplicates, so we just add every
2881 * loose object we find.
2883 static void add_unreachable_loose_objects(void)
2885 for_each_loose_file_in_objdir(get_object_directory(),
2890 static int has_sha1_pack_kept_or_nonlocal(const struct object_id
*oid
)
2892 static struct packed_git
*last_found
= (void *)1;
2893 struct packed_git
*p
;
2895 p
= (last_found
!= (void *)1) ? last_found
:
2896 get_packed_git(the_repository
);
2899 if ((!p
->pack_local
|| p
->pack_keep
||
2900 p
->pack_keep_in_core
) &&
2901 find_pack_entry_one(oid
->hash
, p
)) {
2905 if (p
== last_found
)
2906 p
= get_packed_git(the_repository
);
2909 if (p
== last_found
)
2916 * Store a list of sha1s that are should not be discarded
2917 * because they are either written too recently, or are
2918 * reachable from another object that was.
2920 * This is filled by get_object_list.
2922 static struct oid_array recent_objects
;
2924 static int loosened_object_can_be_discarded(const struct object_id
*oid
,
2927 if (!unpack_unreachable_expiration
)
2929 if (mtime
> unpack_unreachable_expiration
)
2931 if (oid_array_lookup(&recent_objects
, oid
) >= 0)
2936 static void loosen_unused_packed_objects(struct rev_info
*revs
)
2938 struct packed_git
*p
;
2940 struct object_id oid
;
2942 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
2943 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
2946 if (open_pack_index(p
))
2947 die("cannot open pack index");
2949 for (i
= 0; i
< p
->num_objects
; i
++) {
2950 nth_packed_object_oid(&oid
, p
, i
);
2951 if (!packlist_find(&to_pack
, oid
.hash
, NULL
) &&
2952 !has_sha1_pack_kept_or_nonlocal(&oid
) &&
2953 !loosened_object_can_be_discarded(&oid
, p
->mtime
))
2954 if (force_object_loose(&oid
, p
->mtime
))
2955 die("unable to force loose object");
2961 * This tracks any options which pack-reuse code expects to be on, or which a
2962 * reader of the pack might not understand, and which would therefore prevent
2963 * blind reuse of what we have on disk.
2965 static int pack_options_allow_reuse(void)
2967 return pack_to_stdout
&&
2969 !ignore_packed_keep_on_disk
&&
2970 !ignore_packed_keep_in_core
&&
2971 (!local
|| !have_non_local_packs
) &&
2975 static int get_object_list_from_bitmap(struct rev_info
*revs
)
2977 struct bitmap_index
*bitmap_git
;
2978 if (!(bitmap_git
= prepare_bitmap_walk(revs
)))
2981 if (pack_options_allow_reuse() &&
2982 !reuse_partial_packfile_from_bitmap(
2985 &reuse_packfile_objects
,
2986 &reuse_packfile_offset
)) {
2987 assert(reuse_packfile_objects
);
2988 nr_result
+= reuse_packfile_objects
;
2989 display_progress(progress_state
, nr_result
);
2992 traverse_bitmap_commit_list(bitmap_git
, &add_object_entry_from_bitmap
);
2993 free_bitmap_index(bitmap_git
);
2997 static void record_recent_object(struct object
*obj
,
3001 oid_array_append(&recent_objects
, &obj
->oid
);
3004 static void record_recent_commit(struct commit
*commit
, void *data
)
3006 oid_array_append(&recent_objects
, &commit
->object
.oid
);
3009 static void get_object_list(int ac
, const char **av
)
3011 struct rev_info revs
;
3015 init_revisions(&revs
, NULL
);
3016 save_commit_buffer
= 0;
3017 setup_revisions(ac
, av
, &revs
, NULL
);
3019 /* make sure shallows are read */
3020 is_repository_shallow(the_repository
);
3022 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
3023 int len
= strlen(line
);
3024 if (len
&& line
[len
- 1] == '\n')
3029 if (!strcmp(line
, "--not")) {
3030 flags
^= UNINTERESTING
;
3031 write_bitmap_index
= 0;
3034 if (starts_with(line
, "--shallow ")) {
3035 struct object_id oid
;
3036 if (get_oid_hex(line
+ 10, &oid
))
3037 die("not an SHA-1 '%s'", line
+ 10);
3038 register_shallow(the_repository
, &oid
);
3039 use_bitmap_index
= 0;
3042 die("not a rev '%s'", line
);
3044 if (handle_revision_arg(line
, &revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
3045 die("bad revision '%s'", line
);
3048 if (use_bitmap_index
&& !get_object_list_from_bitmap(&revs
))
3051 if (use_delta_islands
)
3052 load_delta_islands();
3054 if (prepare_revision_walk(&revs
))
3055 die("revision walk setup failed");
3056 mark_edges_uninteresting(&revs
, show_edge
);
3058 if (!fn_show_object
)
3059 fn_show_object
= show_object
;
3060 traverse_commit_list_filtered(&filter_options
, &revs
,
3061 show_commit
, fn_show_object
, NULL
,
3064 if (unpack_unreachable_expiration
) {
3065 revs
.ignore_missing_links
= 1;
3066 if (add_unseen_recent_objects_to_traversal(&revs
,
3067 unpack_unreachable_expiration
))
3068 die("unable to add recent objects");
3069 if (prepare_revision_walk(&revs
))
3070 die("revision walk setup failed");
3071 traverse_commit_list(&revs
, record_recent_commit
,
3072 record_recent_object
, NULL
);
3075 if (keep_unreachable
)
3076 add_objects_in_unpacked_packs(&revs
);
3077 if (pack_loose_unreachable
)
3078 add_unreachable_loose_objects();
3079 if (unpack_unreachable
)
3080 loosen_unused_packed_objects(&revs
);
3082 oid_array_clear(&recent_objects
);
3085 static void add_extra_kept_packs(const struct string_list
*names
)
3087 struct packed_git
*p
;
3092 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
3093 const char *name
= basename(p
->pack_name
);
3099 for (i
= 0; i
< names
->nr
; i
++)
3100 if (!fspathcmp(name
, names
->items
[i
].string
))
3103 if (i
< names
->nr
) {
3104 p
->pack_keep_in_core
= 1;
3105 ignore_packed_keep_in_core
= 1;
3111 static int option_parse_index_version(const struct option
*opt
,
3112 const char *arg
, int unset
)
3115 const char *val
= arg
;
3116 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
3117 if (pack_idx_opts
.version
> 2)
3118 die(_("unsupported index version %s"), val
);
3119 if (*c
== ',' && c
[1])
3120 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
3121 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
3122 die(_("bad index version '%s'"), val
);
3126 static int option_parse_unpack_unreachable(const struct option
*opt
,
3127 const char *arg
, int unset
)
3130 unpack_unreachable
= 0;
3131 unpack_unreachable_expiration
= 0;
3134 unpack_unreachable
= 1;
3136 unpack_unreachable_expiration
= approxidate(arg
);
3141 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
3143 int use_internal_rev_list
= 0;
3146 int all_progress_implied
= 0;
3147 struct argv_array rp
= ARGV_ARRAY_INIT
;
3148 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
3149 int rev_list_index
= 0;
3150 struct string_list keep_pack_list
= STRING_LIST_INIT_NODUP
;
3151 struct option pack_objects_options
[] = {
3152 OPT_SET_INT('q', "quiet", &progress
,
3153 N_("do not show progress meter"), 0),
3154 OPT_SET_INT(0, "progress", &progress
,
3155 N_("show progress meter"), 1),
3156 OPT_SET_INT(0, "all-progress", &progress
,
3157 N_("show progress meter during object writing phase"), 2),
3158 OPT_BOOL(0, "all-progress-implied",
3159 &all_progress_implied
,
3160 N_("similar to --all-progress when progress meter is shown")),
3161 { OPTION_CALLBACK
, 0, "index-version", NULL
, N_("version[,offset]"),
3162 N_("write the pack index file in the specified idx format version"),
3163 0, option_parse_index_version
},
3164 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
3165 N_("maximum size of each output pack file")),
3166 OPT_BOOL(0, "local", &local
,
3167 N_("ignore borrowed objects from alternate object store")),
3168 OPT_BOOL(0, "incremental", &incremental
,
3169 N_("ignore packed objects")),
3170 OPT_INTEGER(0, "window", &window
,
3171 N_("limit pack window by objects")),
3172 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
3173 N_("limit pack window by memory in addition to object limit")),
3174 OPT_INTEGER(0, "depth", &depth
,
3175 N_("maximum length of delta chain allowed in the resulting pack")),
3176 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
3177 N_("reuse existing deltas")),
3178 OPT_BOOL(0, "reuse-object", &reuse_object
,
3179 N_("reuse existing objects")),
3180 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
3181 N_("use OFS_DELTA objects")),
3182 OPT_INTEGER(0, "threads", &delta_search_threads
,
3183 N_("use threads when searching for best delta matches")),
3184 OPT_BOOL(0, "non-empty", &non_empty
,
3185 N_("do not create an empty pack output")),
3186 OPT_BOOL(0, "revs", &use_internal_rev_list
,
3187 N_("read revision arguments from standard input")),
3188 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked
,
3189 N_("limit the objects to those that are not yet packed"),
3190 1, PARSE_OPT_NONEG
),
3191 OPT_SET_INT_F(0, "all", &rev_list_all
,
3192 N_("include objects reachable from any reference"),
3193 1, PARSE_OPT_NONEG
),
3194 OPT_SET_INT_F(0, "reflog", &rev_list_reflog
,
3195 N_("include objects referred by reflog entries"),
3196 1, PARSE_OPT_NONEG
),
3197 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index
,
3198 N_("include objects referred to by the index"),
3199 1, PARSE_OPT_NONEG
),
3200 OPT_BOOL(0, "stdout", &pack_to_stdout
,
3201 N_("output pack to stdout")),
3202 OPT_BOOL(0, "include-tag", &include_tag
,
3203 N_("include tag objects that refer to objects to be packed")),
3204 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
3205 N_("keep unreachable objects")),
3206 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
3207 N_("pack loose unreachable objects")),
3208 { OPTION_CALLBACK
, 0, "unpack-unreachable", NULL
, N_("time"),
3209 N_("unpack unreachable objects newer than <time>"),
3210 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
},
3211 OPT_BOOL(0, "thin", &thin
,
3212 N_("create thin packs")),
3213 OPT_BOOL(0, "shallow", &shallow
,
3214 N_("create packs suitable for shallow fetches")),
3215 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk
,
3216 N_("ignore packs that have companion .keep file")),
3217 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list
, N_("name"),
3218 N_("ignore this pack")),
3219 OPT_INTEGER(0, "compression", &pack_compression_level
,
3220 N_("pack compression level")),
3221 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
3222 N_("do not hide commits by grafts"), 0),
3223 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
3224 N_("use a bitmap index if available to speed up counting objects")),
3225 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index
,
3226 N_("write a bitmap index together with the pack index")),
3227 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options
),
3228 { OPTION_CALLBACK
, 0, "missing", NULL
, N_("action"),
3229 N_("handling for missing objects"), PARSE_OPT_NONEG
,
3230 option_parse_missing_action
},
3231 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects
,
3232 N_("do not pack objects in promisor packfiles")),
3233 OPT_BOOL(0, "delta-islands", &use_delta_islands
,
3234 N_("respect islands during delta compression")),
3238 if (DFS_NUM_STATES
> (1 << OE_DFS_STATE_BITS
))
3239 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3241 check_replace_refs
= 0;
3243 reset_pack_idx_option(&pack_idx_opts
);
3244 git_config(git_pack_config
, NULL
);
3246 progress
= isatty(2);
3247 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
3251 base_name
= argv
[0];
3254 if (pack_to_stdout
!= !base_name
|| argc
)
3255 usage_with_options(pack_usage
, pack_objects_options
);
3257 if (depth
>= (1 << OE_DEPTH_BITS
)) {
3258 warning(_("delta chain depth %d is too deep, forcing %d"),
3259 depth
, (1 << OE_DEPTH_BITS
) - 1);
3260 depth
= (1 << OE_DEPTH_BITS
) - 1;
3262 if (cache_max_small_delta_size
>= (1U << OE_Z_DELTA_BITS
)) {
3263 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3264 (1U << OE_Z_DELTA_BITS
) - 1);
3265 cache_max_small_delta_size
= (1U << OE_Z_DELTA_BITS
) - 1;
3268 argv_array_push(&rp
, "pack-objects");
3270 use_internal_rev_list
= 1;
3271 argv_array_push(&rp
, shallow
3272 ? "--objects-edge-aggressive"
3273 : "--objects-edge");
3275 argv_array_push(&rp
, "--objects");
3278 use_internal_rev_list
= 1;
3279 argv_array_push(&rp
, "--all");
3281 if (rev_list_reflog
) {
3282 use_internal_rev_list
= 1;
3283 argv_array_push(&rp
, "--reflog");
3285 if (rev_list_index
) {
3286 use_internal_rev_list
= 1;
3287 argv_array_push(&rp
, "--indexed-objects");
3289 if (rev_list_unpacked
) {
3290 use_internal_rev_list
= 1;
3291 argv_array_push(&rp
, "--unpacked");
3294 if (exclude_promisor_objects
) {
3295 use_internal_rev_list
= 1;
3296 fetch_if_missing
= 0;
3297 argv_array_push(&rp
, "--exclude-promisor-objects");
3299 if (unpack_unreachable
|| keep_unreachable
|| pack_loose_unreachable
)
3300 use_internal_rev_list
= 1;
3304 if (pack_compression_level
== -1)
3305 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
3306 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
3307 die("bad pack compression level %d", pack_compression_level
);
3309 if (!delta_search_threads
) /* --threads=0 means autodetect */
3310 delta_search_threads
= online_cpus();
3313 if (delta_search_threads
!= 1)
3314 warning("no threads support, ignoring --threads");
3316 if (!pack_to_stdout
&& !pack_size_limit
)
3317 pack_size_limit
= pack_size_limit_cfg
;
3318 if (pack_to_stdout
&& pack_size_limit
)
3319 die("--max-pack-size cannot be used to build a pack for transfer.");
3320 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
3321 warning("minimum pack size limit is 1 MiB");
3322 pack_size_limit
= 1024*1024;
3325 if (!pack_to_stdout
&& thin
)
3326 die("--thin cannot be used to build an indexable pack.");
3328 if (keep_unreachable
&& unpack_unreachable
)
3329 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3330 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
3331 unpack_unreachable_expiration
= 0;
3333 if (filter_options
.choice
) {
3334 if (!pack_to_stdout
)
3335 die("cannot use --filter without --stdout.");
3336 use_bitmap_index
= 0;
3340 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3342 * - to produce good pack (with bitmap index not-yet-packed objects are
3343 * packed in suboptimal order).
3345 * - to use more robust pack-generation codepath (avoiding possible
3346 * bugs in bitmap code and possible bitmap index corruption).
3348 if (!pack_to_stdout
)
3349 use_bitmap_index_default
= 0;
3351 if (use_bitmap_index
< 0)
3352 use_bitmap_index
= use_bitmap_index_default
;
3354 /* "hard" reasons not to use bitmaps; these just won't work at all */
3355 if (!use_internal_rev_list
|| (!pack_to_stdout
&& write_bitmap_index
) || is_repository_shallow(the_repository
))
3356 use_bitmap_index
= 0;
3358 if (pack_to_stdout
|| !rev_list_all
)
3359 write_bitmap_index
= 0;
3361 if (use_delta_islands
)
3362 argv_array_push(&rp
, "--topo-order");
3364 if (progress
&& all_progress_implied
)
3367 add_extra_kept_packs(&keep_pack_list
);
3368 if (ignore_packed_keep_on_disk
) {
3369 struct packed_git
*p
;
3370 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
)
3371 if (p
->pack_local
&& p
->pack_keep
)
3373 if (!p
) /* no keep-able packs found */
3374 ignore_packed_keep_on_disk
= 0;
3378 * unlike ignore_packed_keep_on_disk above, we do not
3379 * want to unset "local" based on looking at packs, as
3380 * it also covers non-local objects
3382 struct packed_git
*p
;
3383 for (p
= get_packed_git(the_repository
); p
; p
= p
->next
) {
3384 if (!p
->pack_local
) {
3385 have_non_local_packs
= 1;
3391 prepare_packing_data(&to_pack
);
3394 progress_state
= start_progress(_("Enumerating objects"), 0);
3395 if (!use_internal_rev_list
)
3396 read_object_list_from_stdin();
3398 get_object_list(rp
.argc
, rp
.argv
);
3399 argv_array_clear(&rp
);
3401 cleanup_preferred_base();
3402 if (include_tag
&& nr_result
)
3403 for_each_ref(add_ref_tag
, NULL
);
3404 stop_progress(&progress_state
);
3406 if (non_empty
&& !nr_result
)
3409 prepare_pack(window
, depth
);
3412 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
3413 " reused %"PRIu32
" (delta %"PRIu32
")\n",
3414 written
, written_delta
, reused
, reused_delta
);