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 "oid-array.h"
33 #include "object-store.h"
38 #include "promisor-remote.h"
41 * Objects we are going to pack are collected in the `to_pack` structure.
42 * It contains an array (dynamically expanded) of the object data, and a map
43 * that can resolve SHA1s to their position in the array.
45 static struct packing_data to_pack
;
47 static inline struct object_entry
*oe_delta(
48 const struct packing_data
*pack
,
49 const struct object_entry
*e
)
54 return &pack
->ext_bases
[e
->delta_idx
- 1];
56 return &pack
->objects
[e
->delta_idx
- 1];
59 static inline unsigned long oe_delta_size(struct packing_data
*pack
,
60 const struct object_entry
*e
)
62 if (e
->delta_size_valid
)
63 return e
->delta_size_
;
66 * pack->delta_size[] can't be NULL because oe_set_delta_size()
67 * must have been called when a new delta is saved with
69 * If oe_delta() returns NULL (i.e. default state, which means
70 * delta_size_valid is also false), then the caller must never
71 * call oe_delta_size().
73 return pack
->delta_size
[e
- pack
->objects
];
76 unsigned long oe_get_size_slow(struct packing_data
*pack
,
77 const struct object_entry
*e
);
79 static inline unsigned long oe_size(struct packing_data
*pack
,
80 const struct object_entry
*e
)
85 return oe_get_size_slow(pack
, e
);
88 static inline void oe_set_delta(struct packing_data
*pack
,
89 struct object_entry
*e
,
90 struct object_entry
*delta
)
93 e
->delta_idx
= (delta
- pack
->objects
) + 1;
98 static inline struct object_entry
*oe_delta_sibling(
99 const struct packing_data
*pack
,
100 const struct object_entry
*e
)
102 if (e
->delta_sibling_idx
)
103 return &pack
->objects
[e
->delta_sibling_idx
- 1];
107 static inline struct object_entry
*oe_delta_child(
108 const struct packing_data
*pack
,
109 const struct object_entry
*e
)
111 if (e
->delta_child_idx
)
112 return &pack
->objects
[e
->delta_child_idx
- 1];
116 static inline void oe_set_delta_child(struct packing_data
*pack
,
117 struct object_entry
*e
,
118 struct object_entry
*delta
)
121 e
->delta_child_idx
= (delta
- pack
->objects
) + 1;
123 e
->delta_child_idx
= 0;
126 static inline void oe_set_delta_sibling(struct packing_data
*pack
,
127 struct object_entry
*e
,
128 struct object_entry
*delta
)
131 e
->delta_sibling_idx
= (delta
- pack
->objects
) + 1;
133 e
->delta_sibling_idx
= 0;
136 static inline void oe_set_size(struct packing_data
*pack
,
137 struct object_entry
*e
,
140 if (size
< pack
->oe_size_limit
) {
145 if (oe_get_size_slow(pack
, e
) != size
)
146 BUG("'size' is supposed to be the object size!");
150 static inline void oe_set_delta_size(struct packing_data
*pack
,
151 struct object_entry
*e
,
154 if (size
< pack
->oe_delta_size_limit
) {
155 e
->delta_size_
= size
;
156 e
->delta_size_valid
= 1;
158 packing_data_lock(pack
);
159 if (!pack
->delta_size
)
160 ALLOC_ARRAY(pack
->delta_size
, pack
->nr_alloc
);
161 packing_data_unlock(pack
);
163 pack
->delta_size
[e
- pack
->objects
] = size
;
164 e
->delta_size_valid
= 0;
168 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
169 #define SIZE(obj) oe_size(&to_pack, obj)
170 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
171 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
172 #define DELTA(obj) oe_delta(&to_pack, obj)
173 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
174 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
175 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
176 #define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
177 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
178 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
179 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
181 static const char *pack_usage
[] = {
182 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
183 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
187 static struct pack_idx_entry
**written_list
;
188 static uint32_t nr_result
, nr_written
, nr_seen
;
189 static struct bitmap_index
*bitmap_git
;
190 static uint32_t write_layer
;
192 static int non_empty
;
193 static int reuse_delta
= 1, reuse_object
= 1;
194 static int keep_unreachable
, unpack_unreachable
, include_tag
;
195 static timestamp_t unpack_unreachable_expiration
;
196 static int pack_loose_unreachable
;
198 static int have_non_local_packs
;
199 static int incremental
;
200 static int ignore_packed_keep_on_disk
;
201 static int ignore_packed_keep_in_core
;
202 static int allow_ofs_delta
;
203 static struct pack_idx_option pack_idx_opts
;
204 static const char *base_name
;
205 static int progress
= 1;
206 static int window
= 10;
207 static unsigned long pack_size_limit
;
208 static int depth
= 50;
209 static int delta_search_threads
;
210 static int pack_to_stdout
;
213 static int num_preferred_base
;
214 static struct progress
*progress_state
;
216 static struct packed_git
*reuse_packfile
;
217 static uint32_t reuse_packfile_objects
;
218 static struct bitmap
*reuse_packfile_bitmap
;
220 static int use_bitmap_index_default
= 1;
221 static int use_bitmap_index
= -1;
222 static int allow_pack_reuse
= 1;
224 WRITE_BITMAP_FALSE
= 0,
227 } write_bitmap_index
;
228 static uint16_t write_bitmap_options
= BITMAP_OPT_HASH_CACHE
;
230 static int exclude_promisor_objects
;
232 static int use_delta_islands
;
234 static unsigned long delta_cache_size
= 0;
235 static unsigned long max_delta_cache_size
= DEFAULT_DELTA_CACHE_SIZE
;
236 static unsigned long cache_max_small_delta_size
= 1000;
238 static unsigned long window_memory_limit
= 0;
240 static struct list_objects_filter_options filter_options
;
242 static struct string_list uri_protocols
= STRING_LIST_INIT_NODUP
;
244 enum missing_action
{
245 MA_ERROR
= 0, /* fail if any missing objects are encountered */
246 MA_ALLOW_ANY
, /* silently allow ALL missing objects */
247 MA_ALLOW_PROMISOR
, /* silently allow all missing PROMISOR objects */
249 static enum missing_action arg_missing_action
;
250 static show_object_fn fn_show_object
;
252 struct configured_exclusion
{
253 struct oidmap_entry e
;
257 static struct oidmap configured_exclusions
;
259 static struct oidset excluded_by_config
;
264 static uint32_t written
, written_delta
;
265 static uint32_t reused
, reused_delta
;
270 static struct commit
**indexed_commits
;
271 static unsigned int indexed_commits_nr
;
272 static unsigned int indexed_commits_alloc
;
274 static void index_commit_for_bitmap(struct commit
*commit
)
276 if (indexed_commits_nr
>= indexed_commits_alloc
) {
277 indexed_commits_alloc
= (indexed_commits_alloc
+ 32) * 2;
278 REALLOC_ARRAY(indexed_commits
, indexed_commits_alloc
);
281 indexed_commits
[indexed_commits_nr
++] = commit
;
284 static void *get_delta(struct object_entry
*entry
)
286 unsigned long size
, base_size
, delta_size
;
287 void *buf
, *base_buf
, *delta_buf
;
288 enum object_type type
;
290 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
292 die(_("unable to read %s"), oid_to_hex(&entry
->idx
.oid
));
293 base_buf
= read_object_file(&DELTA(entry
)->idx
.oid
, &type
,
296 die("unable to read %s",
297 oid_to_hex(&DELTA(entry
)->idx
.oid
));
298 delta_buf
= diff_delta(base_buf
, base_size
,
299 buf
, size
, &delta_size
, 0);
301 * We successfully computed this delta once but dropped it for
302 * memory reasons. Something is very wrong if this time we
303 * recompute and create a different delta.
305 if (!delta_buf
|| delta_size
!= DELTA_SIZE(entry
))
306 BUG("delta size changed");
312 static unsigned long do_compress(void **pptr
, unsigned long size
)
316 unsigned long maxsize
;
318 git_deflate_init(&stream
, pack_compression_level
);
319 maxsize
= git_deflate_bound(&stream
, size
);
322 out
= xmalloc(maxsize
);
326 stream
.avail_in
= size
;
327 stream
.next_out
= out
;
328 stream
.avail_out
= maxsize
;
329 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
331 git_deflate_end(&stream
);
334 return stream
.total_out
;
337 static unsigned long write_large_blob_data(struct git_istream
*st
, struct hashfile
*f
,
338 const struct object_id
*oid
)
341 unsigned char ibuf
[1024 * 16];
342 unsigned char obuf
[1024 * 16];
343 unsigned long olen
= 0;
345 git_deflate_init(&stream
, pack_compression_level
);
350 readlen
= read_istream(st
, ibuf
, sizeof(ibuf
));
352 die(_("unable to read %s"), oid_to_hex(oid
));
354 stream
.next_in
= ibuf
;
355 stream
.avail_in
= readlen
;
356 while ((stream
.avail_in
|| readlen
== 0) &&
357 (zret
== Z_OK
|| zret
== Z_BUF_ERROR
)) {
358 stream
.next_out
= obuf
;
359 stream
.avail_out
= sizeof(obuf
);
360 zret
= git_deflate(&stream
, readlen
? 0 : Z_FINISH
);
361 hashwrite(f
, obuf
, stream
.next_out
- obuf
);
362 olen
+= stream
.next_out
- obuf
;
365 die(_("deflate error (%d)"), zret
);
367 if (zret
!= Z_STREAM_END
)
368 die(_("deflate error (%d)"), zret
);
372 git_deflate_end(&stream
);
377 * we are going to reuse the existing object data as is. make
378 * sure it is not corrupt.
380 static int check_pack_inflate(struct packed_git
*p
,
381 struct pack_window
**w_curs
,
384 unsigned long expect
)
387 unsigned char fakebuf
[4096], *in
;
390 memset(&stream
, 0, sizeof(stream
));
391 git_inflate_init(&stream
);
393 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
395 stream
.next_out
= fakebuf
;
396 stream
.avail_out
= sizeof(fakebuf
);
397 st
= git_inflate(&stream
, Z_FINISH
);
398 offset
+= stream
.next_in
- in
;
399 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
400 git_inflate_end(&stream
);
401 return (st
== Z_STREAM_END
&&
402 stream
.total_out
== expect
&&
403 stream
.total_in
== len
) ? 0 : -1;
406 static void copy_pack_data(struct hashfile
*f
,
407 struct packed_git
*p
,
408 struct pack_window
**w_curs
,
416 in
= use_pack(p
, w_curs
, offset
, &avail
);
418 avail
= (unsigned long)len
;
419 hashwrite(f
, in
, avail
);
425 static inline int oe_size_greater_than(struct packing_data
*pack
,
426 const struct object_entry
*lhs
,
430 return lhs
->size_
> rhs
;
431 if (rhs
< pack
->oe_size_limit
) /* rhs < 2^x <= lhs ? */
433 return oe_get_size_slow(pack
, lhs
) > rhs
;
436 /* Return 0 if we will bust the pack-size limit */
437 static unsigned long write_no_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
438 unsigned long limit
, int usable_delta
)
440 unsigned long size
, datalen
;
441 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
442 dheader
[MAX_PACK_OBJECT_HEADER
];
444 enum object_type type
;
446 struct git_istream
*st
= NULL
;
447 const unsigned hashsz
= the_hash_algo
->rawsz
;
450 if (oe_type(entry
) == OBJ_BLOB
&&
451 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
) &&
452 (st
= open_istream(the_repository
, &entry
->idx
.oid
, &type
,
453 &size
, NULL
)) != NULL
)
456 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
458 die(_("unable to read %s"),
459 oid_to_hex(&entry
->idx
.oid
));
462 * make sure no cached delta data remains from a
463 * previous attempt before a pack split occurred.
465 FREE_AND_NULL(entry
->delta_data
);
466 entry
->z_delta_size
= 0;
467 } else if (entry
->delta_data
) {
468 size
= DELTA_SIZE(entry
);
469 buf
= entry
->delta_data
;
470 entry
->delta_data
= NULL
;
471 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
472 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
474 buf
= get_delta(entry
);
475 size
= DELTA_SIZE(entry
);
476 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
477 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
480 if (st
) /* large blob case, just assume we don't compress well */
482 else if (entry
->z_delta_size
)
483 datalen
= entry
->z_delta_size
;
485 datalen
= do_compress(&buf
, size
);
488 * The object header is a byte of 'type' followed by zero or
489 * more bytes of length.
491 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
494 if (type
== OBJ_OFS_DELTA
) {
496 * Deltas with relative base contain an additional
497 * encoding of the relative offset for the delta
498 * base from this object's position in the pack.
500 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
501 unsigned pos
= sizeof(dheader
) - 1;
502 dheader
[pos
] = ofs
& 127;
504 dheader
[--pos
] = 128 | (--ofs
& 127);
505 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
511 hashwrite(f
, header
, hdrlen
);
512 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
513 hdrlen
+= sizeof(dheader
) - pos
;
514 } else if (type
== OBJ_REF_DELTA
) {
516 * Deltas with a base reference contain
517 * additional bytes for the base object ID.
519 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
525 hashwrite(f
, header
, hdrlen
);
526 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
529 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
535 hashwrite(f
, header
, hdrlen
);
538 datalen
= write_large_blob_data(st
, f
, &entry
->idx
.oid
);
541 hashwrite(f
, buf
, datalen
);
545 return hdrlen
+ datalen
;
548 /* Return 0 if we will bust the pack-size limit */
549 static off_t
write_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
550 unsigned long limit
, int usable_delta
)
552 struct packed_git
*p
= IN_PACK(entry
);
553 struct pack_window
*w_curs
= NULL
;
556 enum object_type type
= oe_type(entry
);
558 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
559 dheader
[MAX_PACK_OBJECT_HEADER
];
561 const unsigned hashsz
= the_hash_algo
->rawsz
;
562 unsigned long entry_size
= SIZE(entry
);
565 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
566 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
567 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
570 offset
= entry
->in_pack_offset
;
571 if (offset_to_pack_pos(p
, offset
, &pos
) < 0)
572 die(_("write_reuse_object: could not locate %s, expected at "
573 "offset %"PRIuMAX
" in pack %s"),
574 oid_to_hex(&entry
->idx
.oid
), (uintmax_t)offset
,
576 datalen
= pack_pos_to_offset(p
, pos
+ 1) - offset
;
577 if (!pack_to_stdout
&& p
->index_version
> 1 &&
578 check_pack_crc(p
, &w_curs
, offset
, datalen
,
579 pack_pos_to_index(p
, pos
))) {
580 error(_("bad packed object CRC for %s"),
581 oid_to_hex(&entry
->idx
.oid
));
583 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
586 offset
+= entry
->in_pack_header_size
;
587 datalen
-= entry
->in_pack_header_size
;
589 if (!pack_to_stdout
&& p
->index_version
== 1 &&
590 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry_size
)) {
591 error(_("corrupt packed object for %s"),
592 oid_to_hex(&entry
->idx
.oid
));
594 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
597 if (type
== OBJ_OFS_DELTA
) {
598 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
599 unsigned pos
= sizeof(dheader
) - 1;
600 dheader
[pos
] = ofs
& 127;
602 dheader
[--pos
] = 128 | (--ofs
& 127);
603 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
607 hashwrite(f
, header
, hdrlen
);
608 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
609 hdrlen
+= sizeof(dheader
) - pos
;
611 } else if (type
== OBJ_REF_DELTA
) {
612 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
616 hashwrite(f
, header
, hdrlen
);
617 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
621 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
625 hashwrite(f
, header
, hdrlen
);
627 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
630 return hdrlen
+ datalen
;
633 /* Return 0 if we will bust the pack-size limit */
634 static off_t
write_object(struct hashfile
*f
,
635 struct object_entry
*entry
,
640 int usable_delta
, to_reuse
;
645 /* apply size limit if limited packsize and not first object */
646 if (!pack_size_limit
|| !nr_written
)
648 else if (pack_size_limit
<= write_offset
)
650 * the earlier object did not fit the limit; avoid
651 * mistaking this with unlimited (i.e. limit = 0).
655 limit
= pack_size_limit
- write_offset
;
658 usable_delta
= 0; /* no delta */
659 else if (!pack_size_limit
)
660 usable_delta
= 1; /* unlimited packfile */
661 else if (DELTA(entry
)->idx
.offset
== (off_t
)-1)
662 usable_delta
= 0; /* base was written to another pack */
663 else if (DELTA(entry
)->idx
.offset
)
664 usable_delta
= 1; /* base already exists in this pack */
666 usable_delta
= 0; /* base could end up in another pack */
669 to_reuse
= 0; /* explicit */
670 else if (!IN_PACK(entry
))
671 to_reuse
= 0; /* can't reuse what we don't have */
672 else if (oe_type(entry
) == OBJ_REF_DELTA
||
673 oe_type(entry
) == OBJ_OFS_DELTA
)
674 /* check_object() decided it for us ... */
675 to_reuse
= usable_delta
;
676 /* ... but pack split may override that */
677 else if (oe_type(entry
) != entry
->in_pack_type
)
678 to_reuse
= 0; /* pack has delta which is unusable */
679 else if (DELTA(entry
))
680 to_reuse
= 0; /* we want to pack afresh */
682 to_reuse
= 1; /* we have it in-pack undeltified,
683 * and we do not need to deltify it.
687 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
689 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
697 entry
->idx
.crc32
= crc32_end(f
);
701 enum write_one_status
{
702 WRITE_ONE_SKIP
= -1, /* already written */
703 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
704 WRITE_ONE_WRITTEN
= 1, /* normal */
705 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
708 static enum write_one_status
write_one(struct hashfile
*f
,
709 struct object_entry
*e
,
716 * we set offset to 1 (which is an impossible value) to mark
717 * the fact that this object is involved in "write its base
718 * first before writing a deltified object" recursion.
720 recursing
= (e
->idx
.offset
== 1);
722 warning(_("recursive delta detected for object %s"),
723 oid_to_hex(&e
->idx
.oid
));
724 return WRITE_ONE_RECURSIVE
;
725 } else if (e
->idx
.offset
|| e
->preferred_base
) {
726 /* offset is non zero if object is written already. */
727 return WRITE_ONE_SKIP
;
730 /* if we are deltified, write out base object first. */
732 e
->idx
.offset
= 1; /* now recurse */
733 switch (write_one(f
, DELTA(e
), offset
)) {
734 case WRITE_ONE_RECURSIVE
:
735 /* we cannot depend on this one */
740 case WRITE_ONE_BREAK
:
741 e
->idx
.offset
= recursing
;
742 return WRITE_ONE_BREAK
;
746 e
->idx
.offset
= *offset
;
747 size
= write_object(f
, e
, *offset
);
749 e
->idx
.offset
= recursing
;
750 return WRITE_ONE_BREAK
;
752 written_list
[nr_written
++] = &e
->idx
;
754 /* make sure off_t is sufficiently large not to wrap */
755 if (signed_add_overflows(*offset
, size
))
756 die(_("pack too large for current definition of off_t"));
758 return WRITE_ONE_WRITTEN
;
761 static int mark_tagged(const char *path
, const struct object_id
*oid
, int flag
,
764 struct object_id peeled
;
765 struct object_entry
*entry
= packlist_find(&to_pack
, oid
);
769 if (!peel_iterated_oid(oid
, &peeled
)) {
770 entry
= packlist_find(&to_pack
, &peeled
);
777 static inline unsigned char oe_layer(struct packing_data
*pack
,
778 struct object_entry
*e
)
782 return pack
->layer
[e
- pack
->objects
];
785 static inline void add_to_write_order(struct object_entry
**wo
,
787 struct object_entry
*e
)
789 if (e
->filled
|| oe_layer(&to_pack
, e
) != write_layer
)
795 static void add_descendants_to_write_order(struct object_entry
**wo
,
797 struct object_entry
*e
)
799 int add_to_order
= 1;
802 struct object_entry
*s
;
803 /* add this node... */
804 add_to_write_order(wo
, endp
, e
);
805 /* all its siblings... */
806 for (s
= DELTA_SIBLING(e
); s
; s
= DELTA_SIBLING(s
)) {
807 add_to_write_order(wo
, endp
, s
);
810 /* drop down a level to add left subtree nodes if possible */
811 if (DELTA_CHILD(e
)) {
816 /* our sibling might have some children, it is next */
817 if (DELTA_SIBLING(e
)) {
818 e
= DELTA_SIBLING(e
);
821 /* go back to our parent node */
823 while (e
&& !DELTA_SIBLING(e
)) {
824 /* we're on the right side of a subtree, keep
825 * going up until we can go right again */
829 /* done- we hit our original root node */
832 /* pass it off to sibling at this level */
833 e
= DELTA_SIBLING(e
);
838 static void add_family_to_write_order(struct object_entry
**wo
,
840 struct object_entry
*e
)
842 struct object_entry
*root
;
844 for (root
= e
; DELTA(root
); root
= DELTA(root
))
846 add_descendants_to_write_order(wo
, endp
, root
);
849 static void compute_layer_order(struct object_entry
**wo
, unsigned int *wo_end
)
851 unsigned int i
, last_untagged
;
852 struct object_entry
*objects
= to_pack
.objects
;
854 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
855 if (objects
[i
].tagged
)
857 add_to_write_order(wo
, wo_end
, &objects
[i
]);
862 * Then fill all the tagged tips.
864 for (; i
< to_pack
.nr_objects
; i
++) {
865 if (objects
[i
].tagged
)
866 add_to_write_order(wo
, wo_end
, &objects
[i
]);
870 * And then all remaining commits and tags.
872 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
873 if (oe_type(&objects
[i
]) != OBJ_COMMIT
&&
874 oe_type(&objects
[i
]) != OBJ_TAG
)
876 add_to_write_order(wo
, wo_end
, &objects
[i
]);
880 * And then all the trees.
882 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
883 if (oe_type(&objects
[i
]) != OBJ_TREE
)
885 add_to_write_order(wo
, wo_end
, &objects
[i
]);
889 * Finally all the rest in really tight order
891 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
892 if (!objects
[i
].filled
&& oe_layer(&to_pack
, &objects
[i
]) == write_layer
)
893 add_family_to_write_order(wo
, wo_end
, &objects
[i
]);
897 static struct object_entry
**compute_write_order(void)
899 uint32_t max_layers
= 1;
900 unsigned int i
, wo_end
;
902 struct object_entry
**wo
;
903 struct object_entry
*objects
= to_pack
.objects
;
905 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
906 objects
[i
].tagged
= 0;
907 objects
[i
].filled
= 0;
908 SET_DELTA_CHILD(&objects
[i
], NULL
);
909 SET_DELTA_SIBLING(&objects
[i
], NULL
);
913 * Fully connect delta_child/delta_sibling network.
914 * Make sure delta_sibling is sorted in the original
917 for (i
= to_pack
.nr_objects
; i
> 0;) {
918 struct object_entry
*e
= &objects
[--i
];
921 /* Mark me as the first child */
922 e
->delta_sibling_idx
= DELTA(e
)->delta_child_idx
;
923 SET_DELTA_CHILD(DELTA(e
), e
);
927 * Mark objects that are at the tip of tags.
929 for_each_tag_ref(mark_tagged
, NULL
);
931 if (use_delta_islands
)
932 max_layers
= compute_pack_layers(&to_pack
);
934 ALLOC_ARRAY(wo
, to_pack
.nr_objects
);
937 for (; write_layer
< max_layers
; ++write_layer
)
938 compute_layer_order(wo
, &wo_end
);
940 if (wo_end
!= to_pack
.nr_objects
)
941 die(_("ordered %u objects, expected %"PRIu32
),
942 wo_end
, to_pack
.nr_objects
);
949 * A reused set of objects. All objects in a chunk have the same
950 * relative position in the original packfile and the generated
954 static struct reused_chunk
{
955 /* The offset of the first object of this chunk in the original
958 /* The difference for "original" minus the offset of the first object of
959 * this chunk in the generated packfile. */
962 static int reused_chunks_nr
;
963 static int reused_chunks_alloc
;
965 static void record_reused_object(off_t where
, off_t offset
)
967 if (reused_chunks_nr
&& reused_chunks
[reused_chunks_nr
-1].difference
== offset
)
970 ALLOC_GROW(reused_chunks
, reused_chunks_nr
+ 1,
971 reused_chunks_alloc
);
972 reused_chunks
[reused_chunks_nr
].original
= where
;
973 reused_chunks
[reused_chunks_nr
].difference
= offset
;
978 * Binary search to find the chunk that "where" is in. Note
979 * that we're not looking for an exact match, just the first
980 * chunk that contains it (which implicitly ends at the start
983 static off_t
find_reused_offset(off_t where
)
985 int lo
= 0, hi
= reused_chunks_nr
;
987 int mi
= lo
+ ((hi
- lo
) / 2);
988 if (where
== reused_chunks
[mi
].original
)
989 return reused_chunks
[mi
].difference
;
990 if (where
< reused_chunks
[mi
].original
)
997 * The first chunk starts at zero, so we can't have gone below
1001 return reused_chunks
[lo
-1].difference
;
1004 static void write_reused_pack_one(size_t pos
, struct hashfile
*out
,
1005 struct pack_window
**w_curs
)
1007 off_t offset
, next
, cur
;
1008 enum object_type type
;
1011 offset
= pack_pos_to_offset(reuse_packfile
, pos
);
1012 next
= pack_pos_to_offset(reuse_packfile
, pos
+ 1);
1014 record_reused_object(offset
, offset
- hashfile_total(out
));
1017 type
= unpack_object_header(reuse_packfile
, w_curs
, &cur
, &size
);
1020 if (type
== OBJ_OFS_DELTA
) {
1024 unsigned char header
[MAX_PACK_OBJECT_HEADER
];
1027 base_offset
= get_delta_base(reuse_packfile
, w_curs
, &cur
, type
, offset
);
1028 assert(base_offset
!= 0);
1030 /* Convert to REF_DELTA if we must... */
1031 if (!allow_ofs_delta
) {
1033 struct object_id base_oid
;
1035 if (offset_to_pack_pos(reuse_packfile
, base_offset
, &base_pos
) < 0)
1036 die(_("expected object at offset %"PRIuMAX
" "
1038 (uintmax_t)base_offset
,
1039 reuse_packfile
->pack_name
);
1041 nth_packed_object_id(&base_oid
, reuse_packfile
,
1042 pack_pos_to_index(reuse_packfile
, base_pos
));
1044 len
= encode_in_pack_object_header(header
, sizeof(header
),
1045 OBJ_REF_DELTA
, size
);
1046 hashwrite(out
, header
, len
);
1047 hashwrite(out
, base_oid
.hash
, the_hash_algo
->rawsz
);
1048 copy_pack_data(out
, reuse_packfile
, w_curs
, cur
, next
- cur
);
1052 /* Otherwise see if we need to rewrite the offset... */
1053 fixup
= find_reused_offset(offset
) -
1054 find_reused_offset(base_offset
);
1056 unsigned char ofs_header
[10];
1057 unsigned i
, ofs_len
;
1058 off_t ofs
= offset
- base_offset
- fixup
;
1060 len
= encode_in_pack_object_header(header
, sizeof(header
),
1061 OBJ_OFS_DELTA
, size
);
1063 i
= sizeof(ofs_header
) - 1;
1064 ofs_header
[i
] = ofs
& 127;
1066 ofs_header
[--i
] = 128 | (--ofs
& 127);
1068 ofs_len
= sizeof(ofs_header
) - i
;
1070 hashwrite(out
, header
, len
);
1071 hashwrite(out
, ofs_header
+ sizeof(ofs_header
) - ofs_len
, ofs_len
);
1072 copy_pack_data(out
, reuse_packfile
, w_curs
, cur
, next
- cur
);
1076 /* ...otherwise we have no fixup, and can write it verbatim */
1079 copy_pack_data(out
, reuse_packfile
, w_curs
, offset
, next
- offset
);
1082 static size_t write_reused_pack_verbatim(struct hashfile
*out
,
1083 struct pack_window
**w_curs
)
1087 while (pos
< reuse_packfile_bitmap
->word_alloc
&&
1088 reuse_packfile_bitmap
->words
[pos
] == (eword_t
)~0)
1094 written
= (pos
* BITS_IN_EWORD
);
1095 to_write
= pack_pos_to_offset(reuse_packfile
, written
)
1096 - sizeof(struct pack_header
);
1098 /* We're recording one chunk, not one object. */
1099 record_reused_object(sizeof(struct pack_header
), 0);
1101 copy_pack_data(out
, reuse_packfile
, w_curs
,
1102 sizeof(struct pack_header
), to_write
);
1104 display_progress(progress_state
, written
);
1109 static void write_reused_pack(struct hashfile
*f
)
1113 struct pack_window
*w_curs
= NULL
;
1115 if (allow_ofs_delta
)
1116 i
= write_reused_pack_verbatim(f
, &w_curs
);
1118 for (; i
< reuse_packfile_bitmap
->word_alloc
; ++i
) {
1119 eword_t word
= reuse_packfile_bitmap
->words
[i
];
1120 size_t pos
= (i
* BITS_IN_EWORD
);
1122 for (offset
= 0; offset
< BITS_IN_EWORD
; ++offset
) {
1123 if ((word
>> offset
) == 0)
1126 offset
+= ewah_bit_ctz64(word
>> offset
);
1128 * Can use bit positions directly, even for MIDX
1129 * bitmaps. See comment in try_partial_reuse()
1132 write_reused_pack_one(pos
+ offset
, f
, &w_curs
);
1133 display_progress(progress_state
, ++written
);
1137 unuse_pack(&w_curs
);
1140 static void write_excluded_by_configs(void)
1142 struct oidset_iter iter
;
1143 const struct object_id
*oid
;
1145 oidset_iter_init(&excluded_by_config
, &iter
);
1146 while ((oid
= oidset_iter_next(&iter
))) {
1147 struct configured_exclusion
*ex
=
1148 oidmap_get(&configured_exclusions
, oid
);
1151 BUG("configured exclusion wasn't configured");
1152 write_in_full(1, ex
->pack_hash_hex
, strlen(ex
->pack_hash_hex
));
1153 write_in_full(1, " ", 1);
1154 write_in_full(1, ex
->uri
, strlen(ex
->uri
));
1155 write_in_full(1, "\n", 1);
1159 static const char no_split_warning
[] = N_(
1160 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
1163 static void write_pack_file(void)
1168 uint32_t nr_remaining
= nr_result
;
1169 time_t last_mtime
= 0;
1170 struct object_entry
**write_order
;
1172 if (progress
> pack_to_stdout
)
1173 progress_state
= start_progress(_("Writing objects"), nr_result
);
1174 ALLOC_ARRAY(written_list
, to_pack
.nr_objects
);
1175 write_order
= compute_write_order();
1178 unsigned char hash
[GIT_MAX_RAWSZ
];
1179 char *pack_tmp_name
= NULL
;
1182 f
= hashfd_throughput(1, "<stdout>", progress_state
);
1184 f
= create_tmp_packfile(&pack_tmp_name
);
1186 offset
= write_pack_header(f
, nr_remaining
);
1188 if (reuse_packfile
) {
1189 assert(pack_to_stdout
);
1190 write_reused_pack(f
);
1191 offset
= hashfile_total(f
);
1195 for (; i
< to_pack
.nr_objects
; i
++) {
1196 struct object_entry
*e
= write_order
[i
];
1197 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
1199 display_progress(progress_state
, written
);
1203 * Did we write the wrong # entries in the header?
1204 * If so, rewrite it like in fast-import
1206 if (pack_to_stdout
) {
1207 finalize_hashfile(f
, hash
, CSUM_HASH_IN_STREAM
| CSUM_CLOSE
);
1208 } else if (nr_written
== nr_remaining
) {
1209 finalize_hashfile(f
, hash
, CSUM_HASH_IN_STREAM
| CSUM_FSYNC
| CSUM_CLOSE
);
1211 int fd
= finalize_hashfile(f
, hash
, 0);
1212 fixup_pack_header_footer(fd
, hash
, pack_tmp_name
,
1213 nr_written
, hash
, offset
);
1215 if (write_bitmap_index
) {
1216 if (write_bitmap_index
!= WRITE_BITMAP_QUIET
)
1217 warning(_(no_split_warning
));
1218 write_bitmap_index
= 0;
1222 if (!pack_to_stdout
) {
1224 struct strbuf tmpname
= STRBUF_INIT
;
1225 char *idx_tmp_name
= NULL
;
1228 * Packs are runtime accessed in their mtime
1229 * order since newer packs are more likely to contain
1230 * younger objects. So if we are creating multiple
1231 * packs then we should modify the mtime of later ones
1232 * to preserve this property.
1234 if (stat(pack_tmp_name
, &st
) < 0) {
1235 warning_errno(_("failed to stat %s"), pack_tmp_name
);
1236 } else if (!last_mtime
) {
1237 last_mtime
= st
.st_mtime
;
1240 utb
.actime
= st
.st_atime
;
1241 utb
.modtime
= --last_mtime
;
1242 if (utime(pack_tmp_name
, &utb
) < 0)
1243 warning_errno(_("failed utime() on %s"), pack_tmp_name
);
1246 strbuf_addf(&tmpname
, "%s-%s.", base_name
,
1249 if (write_bitmap_index
) {
1250 bitmap_writer_set_checksum(hash
);
1251 bitmap_writer_build_type_index(
1252 &to_pack
, written_list
, nr_written
);
1255 stage_tmp_packfiles(&tmpname
, pack_tmp_name
,
1256 written_list
, nr_written
,
1257 &pack_idx_opts
, hash
, &idx_tmp_name
);
1259 if (write_bitmap_index
) {
1260 size_t tmpname_len
= tmpname
.len
;
1262 strbuf_addstr(&tmpname
, "bitmap");
1263 stop_progress(&progress_state
);
1265 bitmap_writer_show_progress(progress
);
1266 bitmap_writer_select_commits(indexed_commits
, indexed_commits_nr
, -1);
1267 if (bitmap_writer_build(&to_pack
) < 0)
1268 die(_("failed to write bitmap index"));
1269 bitmap_writer_finish(written_list
, nr_written
,
1270 tmpname
.buf
, write_bitmap_options
);
1271 write_bitmap_index
= 0;
1272 strbuf_setlen(&tmpname
, tmpname_len
);
1275 rename_tmp_packfile_idx(&tmpname
, &idx_tmp_name
);
1278 strbuf_release(&tmpname
);
1279 free(pack_tmp_name
);
1280 puts(hash_to_hex(hash
));
1283 /* mark written objects as written to previous pack */
1284 for (j
= 0; j
< nr_written
; j
++) {
1285 written_list
[j
]->offset
= (off_t
)-1;
1287 nr_remaining
-= nr_written
;
1288 } while (nr_remaining
&& i
< to_pack
.nr_objects
);
1292 stop_progress(&progress_state
);
1293 if (written
!= nr_result
)
1294 die(_("wrote %"PRIu32
" objects while expecting %"PRIu32
),
1295 written
, nr_result
);
1296 trace2_data_intmax("pack-objects", the_repository
,
1297 "write_pack_file/wrote", nr_result
);
1300 static int no_try_delta(const char *path
)
1302 static struct attr_check
*check
;
1305 check
= attr_check_initl("delta", NULL
);
1306 git_check_attr(the_repository
->index
, path
, check
);
1307 if (ATTR_FALSE(check
->items
[0].value
))
1313 * When adding an object, check whether we have already added it
1314 * to our packing list. If so, we can skip. However, if we are
1315 * being asked to excludei t, but the previous mention was to include
1316 * it, make sure to adjust its flags and tweak our numbers accordingly.
1318 * As an optimization, we pass out the index position where we would have
1319 * found the item, since that saves us from having to look it up again a
1320 * few lines later when we want to add the new entry.
1322 static int have_duplicate_entry(const struct object_id
*oid
,
1325 struct object_entry
*entry
;
1327 if (reuse_packfile_bitmap
&&
1328 bitmap_walk_contains(bitmap_git
, reuse_packfile_bitmap
, oid
))
1331 entry
= packlist_find(&to_pack
, oid
);
1336 if (!entry
->preferred_base
)
1338 entry
->preferred_base
= 1;
1344 static int want_found_object(const struct object_id
*oid
, int exclude
,
1345 struct packed_git
*p
)
1353 * When asked to do --local (do not include an object that appears in a
1354 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1355 * an object that appears in a pack marked with .keep), finding a pack
1356 * that matches the criteria is sufficient for us to decide to omit it.
1357 * However, even if this pack does not satisfy the criteria, we need to
1358 * make sure no copy of this object appears in _any_ pack that makes us
1359 * to omit the object, so we need to check all the packs.
1361 * We can however first check whether these options can possibly matter;
1362 * if they do not matter we know we want the object in generated pack.
1363 * Otherwise, we signal "-1" at the end to tell the caller that we do
1364 * not know either way, and it needs to check more packs.
1368 * Objects in packs borrowed from elsewhere are discarded regardless of
1369 * if they appear in other packs that weren't borrowed.
1371 if (local
&& !p
->pack_local
)
1375 * Then handle .keep first, as we have a fast(er) path there.
1377 if (ignore_packed_keep_on_disk
|| ignore_packed_keep_in_core
) {
1379 * Set the flags for the kept-pack cache to be the ones we want
1382 * That is, if we are ignoring objects in on-disk keep packs,
1383 * then we want to search through the on-disk keep and ignore
1387 if (ignore_packed_keep_on_disk
)
1388 flags
|= ON_DISK_KEEP_PACKS
;
1389 if (ignore_packed_keep_in_core
)
1390 flags
|= IN_CORE_KEEP_PACKS
;
1392 if (ignore_packed_keep_on_disk
&& p
->pack_keep
)
1394 if (ignore_packed_keep_in_core
&& p
->pack_keep_in_core
)
1396 if (has_object_kept_pack(oid
, flags
))
1401 * At this point we know definitively that either we don't care about
1402 * keep-packs, or the object is not in one. Keep checking other
1405 if (!local
|| !have_non_local_packs
)
1408 /* we don't know yet; keep looking for more packs */
1412 static int want_object_in_pack_one(struct packed_git
*p
,
1413 const struct object_id
*oid
,
1415 struct packed_git
**found_pack
,
1416 off_t
*found_offset
)
1420 if (p
== *found_pack
)
1421 offset
= *found_offset
;
1423 offset
= find_pack_entry_one(oid
->hash
, p
);
1427 if (!is_pack_valid(p
))
1429 *found_offset
= offset
;
1432 return want_found_object(oid
, exclude
, p
);
1438 * Check whether we want the object in the pack (e.g., we do not want
1439 * objects found in non-local stores if the "--local" option was used).
1441 * If the caller already knows an existing pack it wants to take the object
1442 * from, that is passed in *found_pack and *found_offset; otherwise this
1443 * function finds if there is any pack that has the object and returns the pack
1444 * and its offset in these variables.
1446 static int want_object_in_pack(const struct object_id
*oid
,
1448 struct packed_git
**found_pack
,
1449 off_t
*found_offset
)
1452 struct list_head
*pos
;
1453 struct multi_pack_index
*m
;
1455 if (!exclude
&& local
&& has_loose_object_nonlocal(oid
))
1459 * If we already know the pack object lives in, start checks from that
1460 * pack - in the usual case when neither --local was given nor .keep files
1461 * are present we will determine the answer right now.
1464 want
= want_found_object(oid
, exclude
, *found_pack
);
1469 for (m
= get_multi_pack_index(the_repository
); m
; m
= m
->next
) {
1470 struct pack_entry e
;
1471 if (fill_midx_entry(the_repository
, oid
, &e
, m
)) {
1472 want
= want_object_in_pack_one(e
.p
, oid
, exclude
, found_pack
, found_offset
);
1478 list_for_each(pos
, get_packed_git_mru(the_repository
)) {
1479 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1480 want
= want_object_in_pack_one(p
, oid
, exclude
, found_pack
, found_offset
);
1481 if (!exclude
&& want
> 0)
1483 get_packed_git_mru(the_repository
));
1488 if (uri_protocols
.nr
) {
1489 struct configured_exclusion
*ex
=
1490 oidmap_get(&configured_exclusions
, oid
);
1495 for (i
= 0; i
< uri_protocols
.nr
; i
++) {
1496 if (skip_prefix(ex
->uri
,
1497 uri_protocols
.items
[i
].string
,
1500 oidset_insert(&excluded_by_config
, oid
);
1510 static void create_object_entry(const struct object_id
*oid
,
1511 enum object_type type
,
1515 struct packed_git
*found_pack
,
1518 struct object_entry
*entry
;
1520 entry
= packlist_alloc(&to_pack
, oid
);
1522 oe_set_type(entry
, type
);
1524 entry
->preferred_base
= 1;
1528 oe_set_in_pack(&to_pack
, entry
, found_pack
);
1529 entry
->in_pack_offset
= found_offset
;
1532 entry
->no_try_delta
= no_try_delta
;
1535 static const char no_closure_warning
[] = N_(
1536 "disabling bitmap writing, as some objects are not being packed"
1539 static int add_object_entry(const struct object_id
*oid
, enum object_type type
,
1540 const char *name
, int exclude
)
1542 struct packed_git
*found_pack
= NULL
;
1543 off_t found_offset
= 0;
1545 display_progress(progress_state
, ++nr_seen
);
1547 if (have_duplicate_entry(oid
, exclude
))
1550 if (!want_object_in_pack(oid
, exclude
, &found_pack
, &found_offset
)) {
1551 /* The pack is missing an object, so it will not have closure */
1552 if (write_bitmap_index
) {
1553 if (write_bitmap_index
!= WRITE_BITMAP_QUIET
)
1554 warning(_(no_closure_warning
));
1555 write_bitmap_index
= 0;
1560 create_object_entry(oid
, type
, pack_name_hash(name
),
1561 exclude
, name
&& no_try_delta(name
),
1562 found_pack
, found_offset
);
1566 static int add_object_entry_from_bitmap(const struct object_id
*oid
,
1567 enum object_type type
,
1568 int flags
, uint32_t name_hash
,
1569 struct packed_git
*pack
, off_t offset
)
1571 display_progress(progress_state
, ++nr_seen
);
1573 if (have_duplicate_entry(oid
, 0))
1576 if (!want_object_in_pack(oid
, 0, &pack
, &offset
))
1579 create_object_entry(oid
, type
, name_hash
, 0, 0, pack
, offset
);
1583 struct pbase_tree_cache
{
1584 struct object_id oid
;
1588 unsigned long tree_size
;
1591 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1592 static int pbase_tree_cache_ix(const struct object_id
*oid
)
1594 return oid
->hash
[0] % ARRAY_SIZE(pbase_tree_cache
);
1596 static int pbase_tree_cache_ix_incr(int ix
)
1598 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1601 static struct pbase_tree
{
1602 struct pbase_tree
*next
;
1603 /* This is a phony "cache" entry; we are not
1604 * going to evict it or find it through _get()
1605 * mechanism -- this is for the toplevel node that
1606 * would almost always change with any commit.
1608 struct pbase_tree_cache pcache
;
1611 static struct pbase_tree_cache
*pbase_tree_get(const struct object_id
*oid
)
1613 struct pbase_tree_cache
*ent
, *nent
;
1616 enum object_type type
;
1618 int my_ix
= pbase_tree_cache_ix(oid
);
1619 int available_ix
= -1;
1621 /* pbase-tree-cache acts as a limited hashtable.
1622 * your object will be found at your index or within a few
1623 * slots after that slot if it is cached.
1625 for (neigh
= 0; neigh
< 8; neigh
++) {
1626 ent
= pbase_tree_cache
[my_ix
];
1627 if (ent
&& oideq(&ent
->oid
, oid
)) {
1631 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1632 ((0 <= available_ix
) &&
1633 (!ent
&& pbase_tree_cache
[available_ix
])))
1634 available_ix
= my_ix
;
1637 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1640 /* Did not find one. Either we got a bogus request or
1641 * we need to read and perhaps cache.
1643 data
= read_object_file(oid
, &type
, &size
);
1646 if (type
!= OBJ_TREE
) {
1651 /* We need to either cache or return a throwaway copy */
1653 if (available_ix
< 0)
1656 ent
= pbase_tree_cache
[available_ix
];
1657 my_ix
= available_ix
;
1661 nent
= xmalloc(sizeof(*nent
));
1662 nent
->temporary
= (available_ix
< 0);
1665 /* evict and reuse */
1666 free(ent
->tree_data
);
1669 oidcpy(&nent
->oid
, oid
);
1670 nent
->tree_data
= data
;
1671 nent
->tree_size
= size
;
1673 if (!nent
->temporary
)
1674 pbase_tree_cache
[my_ix
] = nent
;
1678 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1680 if (!cache
->temporary
) {
1684 free(cache
->tree_data
);
1688 static int name_cmp_len(const char *name
)
1691 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
1696 static void add_pbase_object(struct tree_desc
*tree
,
1699 const char *fullname
)
1701 struct name_entry entry
;
1704 while (tree_entry(tree
,&entry
)) {
1705 if (S_ISGITLINK(entry
.mode
))
1707 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1708 memcmp(name
, entry
.path
, cmplen
);
1713 if (name
[cmplen
] != '/') {
1714 add_object_entry(&entry
.oid
,
1715 object_type(entry
.mode
),
1719 if (S_ISDIR(entry
.mode
)) {
1720 struct tree_desc sub
;
1721 struct pbase_tree_cache
*tree
;
1722 const char *down
= name
+cmplen
+1;
1723 int downlen
= name_cmp_len(down
);
1725 tree
= pbase_tree_get(&entry
.oid
);
1728 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1730 add_pbase_object(&sub
, down
, downlen
, fullname
);
1731 pbase_tree_put(tree
);
1736 static unsigned *done_pbase_paths
;
1737 static int done_pbase_paths_num
;
1738 static int done_pbase_paths_alloc
;
1739 static int done_pbase_path_pos(unsigned hash
)
1742 int hi
= done_pbase_paths_num
;
1744 int mi
= lo
+ (hi
- lo
) / 2;
1745 if (done_pbase_paths
[mi
] == hash
)
1747 if (done_pbase_paths
[mi
] < hash
)
1755 static int check_pbase_path(unsigned hash
)
1757 int pos
= done_pbase_path_pos(hash
);
1761 ALLOC_GROW(done_pbase_paths
,
1762 done_pbase_paths_num
+ 1,
1763 done_pbase_paths_alloc
);
1764 done_pbase_paths_num
++;
1765 if (pos
< done_pbase_paths_num
)
1766 MOVE_ARRAY(done_pbase_paths
+ pos
+ 1, done_pbase_paths
+ pos
,
1767 done_pbase_paths_num
- pos
- 1);
1768 done_pbase_paths
[pos
] = hash
;
1772 static void add_preferred_base_object(const char *name
)
1774 struct pbase_tree
*it
;
1776 unsigned hash
= pack_name_hash(name
);
1778 if (!num_preferred_base
|| check_pbase_path(hash
))
1781 cmplen
= name_cmp_len(name
);
1782 for (it
= pbase_tree
; it
; it
= it
->next
) {
1784 add_object_entry(&it
->pcache
.oid
, OBJ_TREE
, NULL
, 1);
1787 struct tree_desc tree
;
1788 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1789 add_pbase_object(&tree
, name
, cmplen
, name
);
1794 static void add_preferred_base(struct object_id
*oid
)
1796 struct pbase_tree
*it
;
1799 struct object_id tree_oid
;
1801 if (window
<= num_preferred_base
++)
1804 data
= read_object_with_reference(the_repository
, oid
,
1805 tree_type
, &size
, &tree_oid
);
1809 for (it
= pbase_tree
; it
; it
= it
->next
) {
1810 if (oideq(&it
->pcache
.oid
, &tree_oid
)) {
1816 CALLOC_ARRAY(it
, 1);
1817 it
->next
= pbase_tree
;
1820 oidcpy(&it
->pcache
.oid
, &tree_oid
);
1821 it
->pcache
.tree_data
= data
;
1822 it
->pcache
.tree_size
= size
;
1825 static void cleanup_preferred_base(void)
1827 struct pbase_tree
*it
;
1833 struct pbase_tree
*tmp
= it
;
1835 free(tmp
->pcache
.tree_data
);
1839 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1840 if (!pbase_tree_cache
[i
])
1842 free(pbase_tree_cache
[i
]->tree_data
);
1843 FREE_AND_NULL(pbase_tree_cache
[i
]);
1846 FREE_AND_NULL(done_pbase_paths
);
1847 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1851 * Return 1 iff the object specified by "delta" can be sent
1852 * literally as a delta against the base in "base_sha1". If
1853 * so, then *base_out will point to the entry in our packing
1854 * list, or NULL if we must use the external-base list.
1856 * Depth value does not matter - find_deltas() will
1857 * never consider reused delta as the base object to
1858 * deltify other objects against, in order to avoid
1861 static int can_reuse_delta(const struct object_id
*base_oid
,
1862 struct object_entry
*delta
,
1863 struct object_entry
**base_out
)
1865 struct object_entry
*base
;
1868 * First see if we're already sending the base (or it's explicitly in
1869 * our "excluded" list).
1871 base
= packlist_find(&to_pack
, base_oid
);
1873 if (!in_same_island(&delta
->idx
.oid
, &base
->idx
.oid
))
1880 * Otherwise, reachability bitmaps may tell us if the receiver has it,
1881 * even if it was buried too deep in history to make it into the
1884 if (thin
&& bitmap_has_oid_in_uninteresting(bitmap_git
, base_oid
)) {
1885 if (use_delta_islands
) {
1886 if (!in_same_island(&delta
->idx
.oid
, base_oid
))
1896 static void prefetch_to_pack(uint32_t object_index_start
) {
1897 struct oid_array to_fetch
= OID_ARRAY_INIT
;
1900 for (i
= object_index_start
; i
< to_pack
.nr_objects
; i
++) {
1901 struct object_entry
*entry
= to_pack
.objects
+ i
;
1903 if (!oid_object_info_extended(the_repository
,
1906 OBJECT_INFO_FOR_PREFETCH
))
1908 oid_array_append(&to_fetch
, &entry
->idx
.oid
);
1910 promisor_remote_get_direct(the_repository
,
1911 to_fetch
.oid
, to_fetch
.nr
);
1912 oid_array_clear(&to_fetch
);
1915 static void check_object(struct object_entry
*entry
, uint32_t object_index
)
1917 unsigned long canonical_size
;
1918 enum object_type type
;
1919 struct object_info oi
= {.typep
= &type
, .sizep
= &canonical_size
};
1921 if (IN_PACK(entry
)) {
1922 struct packed_git
*p
= IN_PACK(entry
);
1923 struct pack_window
*w_curs
= NULL
;
1925 struct object_id base_ref
;
1926 struct object_entry
*base_entry
;
1927 unsigned long used
, used_0
;
1928 unsigned long avail
;
1930 unsigned char *buf
, c
;
1931 enum object_type type
;
1932 unsigned long in_pack_size
;
1934 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1937 * We want in_pack_type even if we do not reuse delta
1938 * since non-delta representations could still be reused.
1940 used
= unpack_object_header_buffer(buf
, avail
,
1947 BUG("invalid type %d", type
);
1948 entry
->in_pack_type
= type
;
1951 * Determine if this is a delta and if so whether we can
1952 * reuse it or not. Otherwise let's find out as cheaply as
1953 * possible what the actual type and size for this object is.
1955 switch (entry
->in_pack_type
) {
1957 /* Not a delta hence we've already got all we need. */
1958 oe_set_type(entry
, entry
->in_pack_type
);
1959 SET_SIZE(entry
, in_pack_size
);
1960 entry
->in_pack_header_size
= used
;
1961 if (oe_type(entry
) < OBJ_COMMIT
|| oe_type(entry
) > OBJ_BLOB
)
1963 unuse_pack(&w_curs
);
1966 if (reuse_delta
&& !entry
->preferred_base
) {
1968 use_pack(p
, &w_curs
,
1969 entry
->in_pack_offset
+ used
,
1973 entry
->in_pack_header_size
= used
+ the_hash_algo
->rawsz
;
1976 buf
= use_pack(p
, &w_curs
,
1977 entry
->in_pack_offset
+ used
, NULL
);
1983 if (!ofs
|| MSB(ofs
, 7)) {
1984 error(_("delta base offset overflow in pack for %s"),
1985 oid_to_hex(&entry
->idx
.oid
));
1989 ofs
= (ofs
<< 7) + (c
& 127);
1991 ofs
= entry
->in_pack_offset
- ofs
;
1992 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1993 error(_("delta base offset out of bound for %s"),
1994 oid_to_hex(&entry
->idx
.oid
));
1997 if (reuse_delta
&& !entry
->preferred_base
) {
1999 if (offset_to_pack_pos(p
, ofs
, &pos
) < 0)
2001 if (!nth_packed_object_id(&base_ref
, p
,
2002 pack_pos_to_index(p
, pos
)))
2005 entry
->in_pack_header_size
= used
+ used_0
;
2010 can_reuse_delta(&base_ref
, entry
, &base_entry
)) {
2011 oe_set_type(entry
, entry
->in_pack_type
);
2012 SET_SIZE(entry
, in_pack_size
); /* delta size */
2013 SET_DELTA_SIZE(entry
, in_pack_size
);
2016 SET_DELTA(entry
, base_entry
);
2017 entry
->delta_sibling_idx
= base_entry
->delta_child_idx
;
2018 SET_DELTA_CHILD(base_entry
, entry
);
2020 SET_DELTA_EXT(entry
, &base_ref
);
2023 unuse_pack(&w_curs
);
2027 if (oe_type(entry
)) {
2031 * This must be a delta and we already know what the
2032 * final object type is. Let's extract the actual
2033 * object size from the delta header.
2035 delta_pos
= entry
->in_pack_offset
+ entry
->in_pack_header_size
;
2036 canonical_size
= get_size_from_delta(p
, &w_curs
, delta_pos
);
2037 if (canonical_size
== 0)
2039 SET_SIZE(entry
, canonical_size
);
2040 unuse_pack(&w_curs
);
2045 * No choice but to fall back to the recursive delta walk
2046 * with oid_object_info() to find about the object type
2050 unuse_pack(&w_curs
);
2053 if (oid_object_info_extended(the_repository
, &entry
->idx
.oid
, &oi
,
2054 OBJECT_INFO_SKIP_FETCH_OBJECT
| OBJECT_INFO_LOOKUP_REPLACE
) < 0) {
2055 if (has_promisor_remote()) {
2056 prefetch_to_pack(object_index
);
2057 if (oid_object_info_extended(the_repository
, &entry
->idx
.oid
, &oi
,
2058 OBJECT_INFO_SKIP_FETCH_OBJECT
| OBJECT_INFO_LOOKUP_REPLACE
) < 0)
2064 oe_set_type(entry
, type
);
2065 if (entry
->type_valid
) {
2066 SET_SIZE(entry
, canonical_size
);
2069 * Bad object type is checked in prepare_pack(). This is
2070 * to permit a missing preferred base object to be ignored
2071 * as a preferred base. Doing so can result in a larger
2072 * pack file, but the transfer will still take place.
2077 static int pack_offset_sort(const void *_a
, const void *_b
)
2079 const struct object_entry
*a
= *(struct object_entry
**)_a
;
2080 const struct object_entry
*b
= *(struct object_entry
**)_b
;
2081 const struct packed_git
*a_in_pack
= IN_PACK(a
);
2082 const struct packed_git
*b_in_pack
= IN_PACK(b
);
2084 /* avoid filesystem trashing with loose objects */
2085 if (!a_in_pack
&& !b_in_pack
)
2086 return oidcmp(&a
->idx
.oid
, &b
->idx
.oid
);
2088 if (a_in_pack
< b_in_pack
)
2090 if (a_in_pack
> b_in_pack
)
2092 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
2093 (a
->in_pack_offset
> b
->in_pack_offset
);
2097 * Drop an on-disk delta we were planning to reuse. Naively, this would
2098 * just involve blanking out the "delta" field, but we have to deal
2099 * with some extra book-keeping:
2101 * 1. Removing ourselves from the delta_sibling linked list.
2103 * 2. Updating our size/type to the non-delta representation. These were
2104 * either not recorded initially (size) or overwritten with the delta type
2105 * (type) when check_object() decided to reuse the delta.
2107 * 3. Resetting our delta depth, as we are now a base object.
2109 static void drop_reused_delta(struct object_entry
*entry
)
2111 unsigned *idx
= &to_pack
.objects
[entry
->delta_idx
- 1].delta_child_idx
;
2112 struct object_info oi
= OBJECT_INFO_INIT
;
2113 enum object_type type
;
2117 struct object_entry
*oe
= &to_pack
.objects
[*idx
- 1];
2120 *idx
= oe
->delta_sibling_idx
;
2122 idx
= &oe
->delta_sibling_idx
;
2124 SET_DELTA(entry
, NULL
);
2129 if (packed_object_info(the_repository
, IN_PACK(entry
), entry
->in_pack_offset
, &oi
) < 0) {
2131 * We failed to get the info from this pack for some reason;
2132 * fall back to oid_object_info, which may find another copy.
2133 * And if that fails, the error will be recorded in oe_type(entry)
2134 * and dealt with in prepare_pack().
2137 oid_object_info(the_repository
, &entry
->idx
.oid
, &size
));
2139 oe_set_type(entry
, type
);
2141 SET_SIZE(entry
, size
);
2145 * Follow the chain of deltas from this entry onward, throwing away any links
2146 * that cause us to hit a cycle (as determined by the DFS state flags in
2149 * We also detect too-long reused chains that would violate our --depth
2152 static void break_delta_chains(struct object_entry
*entry
)
2155 * The actual depth of each object we will write is stored as an int,
2156 * as it cannot exceed our int "depth" limit. But before we break
2157 * changes based no that limit, we may potentially go as deep as the
2158 * number of objects, which is elsewhere bounded to a uint32_t.
2160 uint32_t total_depth
;
2161 struct object_entry
*cur
, *next
;
2163 for (cur
= entry
, total_depth
= 0;
2165 cur
= DELTA(cur
), total_depth
++) {
2166 if (cur
->dfs_state
== DFS_DONE
) {
2168 * We've already seen this object and know it isn't
2169 * part of a cycle. We do need to append its depth
2172 total_depth
+= cur
->depth
;
2177 * We break cycles before looping, so an ACTIVE state (or any
2178 * other cruft which made its way into the state variable)
2181 if (cur
->dfs_state
!= DFS_NONE
)
2182 BUG("confusing delta dfs state in first pass: %d",
2186 * Now we know this is the first time we've seen the object. If
2187 * it's not a delta, we're done traversing, but we'll mark it
2188 * done to save time on future traversals.
2191 cur
->dfs_state
= DFS_DONE
;
2196 * Mark ourselves as active and see if the next step causes
2197 * us to cycle to another active object. It's important to do
2198 * this _before_ we loop, because it impacts where we make the
2199 * cut, and thus how our total_depth counter works.
2200 * E.g., We may see a partial loop like:
2202 * A -> B -> C -> D -> B
2204 * Cutting B->C breaks the cycle. But now the depth of A is
2205 * only 1, and our total_depth counter is at 3. The size of the
2206 * error is always one less than the size of the cycle we
2207 * broke. Commits C and D were "lost" from A's chain.
2209 * If we instead cut D->B, then the depth of A is correct at 3.
2210 * We keep all commits in the chain that we examined.
2212 cur
->dfs_state
= DFS_ACTIVE
;
2213 if (DELTA(cur
)->dfs_state
== DFS_ACTIVE
) {
2214 drop_reused_delta(cur
);
2215 cur
->dfs_state
= DFS_DONE
;
2221 * And now that we've gone all the way to the bottom of the chain, we
2222 * need to clear the active flags and set the depth fields as
2223 * appropriate. Unlike the loop above, which can quit when it drops a
2224 * delta, we need to keep going to look for more depth cuts. So we need
2225 * an extra "next" pointer to keep going after we reset cur->delta.
2227 for (cur
= entry
; cur
; cur
= next
) {
2231 * We should have a chain of zero or more ACTIVE states down to
2232 * a final DONE. We can quit after the DONE, because either it
2233 * has no bases, or we've already handled them in a previous
2236 if (cur
->dfs_state
== DFS_DONE
)
2238 else if (cur
->dfs_state
!= DFS_ACTIVE
)
2239 BUG("confusing delta dfs state in second pass: %d",
2243 * If the total_depth is more than depth, then we need to snip
2244 * the chain into two or more smaller chains that don't exceed
2245 * the maximum depth. Most of the resulting chains will contain
2246 * (depth + 1) entries (i.e., depth deltas plus one base), and
2247 * the last chain (i.e., the one containing entry) will contain
2248 * whatever entries are left over, namely
2249 * (total_depth % (depth + 1)) of them.
2251 * Since we are iterating towards decreasing depth, we need to
2252 * decrement total_depth as we go, and we need to write to the
2253 * entry what its final depth will be after all of the
2254 * snipping. Since we're snipping into chains of length (depth
2255 * + 1) entries, the final depth of an entry will be its
2256 * original depth modulo (depth + 1). Any time we encounter an
2257 * entry whose final depth is supposed to be zero, we snip it
2258 * from its delta base, thereby making it so.
2260 cur
->depth
= (total_depth
--) % (depth
+ 1);
2262 drop_reused_delta(cur
);
2264 cur
->dfs_state
= DFS_DONE
;
2268 static void get_object_details(void)
2271 struct object_entry
**sorted_by_offset
;
2274 progress_state
= start_progress(_("Counting objects"),
2275 to_pack
.nr_objects
);
2277 CALLOC_ARRAY(sorted_by_offset
, to_pack
.nr_objects
);
2278 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
2279 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
2280 QSORT(sorted_by_offset
, to_pack
.nr_objects
, pack_offset_sort
);
2282 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2283 struct object_entry
*entry
= sorted_by_offset
[i
];
2284 check_object(entry
, i
);
2285 if (entry
->type_valid
&&
2286 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
))
2287 entry
->no_try_delta
= 1;
2288 display_progress(progress_state
, i
+ 1);
2290 stop_progress(&progress_state
);
2293 * This must happen in a second pass, since we rely on the delta
2294 * information for the whole list being completed.
2296 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
2297 break_delta_chains(&to_pack
.objects
[i
]);
2299 free(sorted_by_offset
);
2303 * We search for deltas in a list sorted by type, by filename hash, and then
2304 * by size, so that we see progressively smaller and smaller files.
2305 * That's because we prefer deltas to be from the bigger file
2306 * to the smaller -- deletes are potentially cheaper, but perhaps
2307 * more importantly, the bigger file is likely the more recent
2308 * one. The deepest deltas are therefore the oldest objects which are
2309 * less susceptible to be accessed often.
2311 static int type_size_sort(const void *_a
, const void *_b
)
2313 const struct object_entry
*a
= *(struct object_entry
**)_a
;
2314 const struct object_entry
*b
= *(struct object_entry
**)_b
;
2315 const enum object_type a_type
= oe_type(a
);
2316 const enum object_type b_type
= oe_type(b
);
2317 const unsigned long a_size
= SIZE(a
);
2318 const unsigned long b_size
= SIZE(b
);
2320 if (a_type
> b_type
)
2322 if (a_type
< b_type
)
2324 if (a
->hash
> b
->hash
)
2326 if (a
->hash
< b
->hash
)
2328 if (a
->preferred_base
> b
->preferred_base
)
2330 if (a
->preferred_base
< b
->preferred_base
)
2332 if (use_delta_islands
) {
2333 const int island_cmp
= island_delta_cmp(&a
->idx
.oid
, &b
->idx
.oid
);
2337 if (a_size
> b_size
)
2339 if (a_size
< b_size
)
2341 return a
< b
? -1 : (a
> b
); /* newest first */
2345 struct object_entry
*entry
;
2347 struct delta_index
*index
;
2351 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
2352 unsigned long delta_size
)
2354 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
2357 if (delta_size
< cache_max_small_delta_size
)
2360 /* cache delta, if objects are large enough compared to delta size */
2361 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
2367 /* Protect delta_cache_size */
2368 static pthread_mutex_t cache_mutex
;
2369 #define cache_lock() pthread_mutex_lock(&cache_mutex)
2370 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
2373 * Protect object list partitioning (e.g. struct thread_param) and
2376 static pthread_mutex_t progress_mutex
;
2377 #define progress_lock() pthread_mutex_lock(&progress_mutex)
2378 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
2381 * Access to struct object_entry is unprotected since each thread owns
2382 * a portion of the main object list. Just don't access object entries
2383 * ahead in the list because they can be stolen and would need
2384 * progress_mutex for protection.
2387 static inline int oe_size_less_than(struct packing_data
*pack
,
2388 const struct object_entry
*lhs
,
2391 if (lhs
->size_valid
)
2392 return lhs
->size_
< rhs
;
2393 if (rhs
< pack
->oe_size_limit
) /* rhs < 2^x <= lhs ? */
2395 return oe_get_size_slow(pack
, lhs
) < rhs
;
2398 static inline void oe_set_tree_depth(struct packing_data
*pack
,
2399 struct object_entry
*e
,
2400 unsigned int tree_depth
)
2402 if (!pack
->tree_depth
)
2403 CALLOC_ARRAY(pack
->tree_depth
, pack
->nr_alloc
);
2404 pack
->tree_depth
[e
- pack
->objects
] = tree_depth
;
2408 * Return the size of the object without doing any delta
2409 * reconstruction (so non-deltas are true object sizes, but deltas
2410 * return the size of the delta data).
2412 unsigned long oe_get_size_slow(struct packing_data
*pack
,
2413 const struct object_entry
*e
)
2415 struct packed_git
*p
;
2416 struct pack_window
*w_curs
;
2418 enum object_type type
;
2419 unsigned long used
, avail
, size
;
2421 if (e
->type_
!= OBJ_OFS_DELTA
&& e
->type_
!= OBJ_REF_DELTA
) {
2422 packing_data_lock(&to_pack
);
2423 if (oid_object_info(the_repository
, &e
->idx
.oid
, &size
) < 0)
2424 die(_("unable to get size of %s"),
2425 oid_to_hex(&e
->idx
.oid
));
2426 packing_data_unlock(&to_pack
);
2430 p
= oe_in_pack(pack
, e
);
2432 BUG("when e->type is a delta, it must belong to a pack");
2434 packing_data_lock(&to_pack
);
2436 buf
= use_pack(p
, &w_curs
, e
->in_pack_offset
, &avail
);
2437 used
= unpack_object_header_buffer(buf
, avail
, &type
, &size
);
2439 die(_("unable to parse object header of %s"),
2440 oid_to_hex(&e
->idx
.oid
));
2442 unuse_pack(&w_curs
);
2443 packing_data_unlock(&to_pack
);
2447 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
2448 unsigned max_depth
, unsigned long *mem_usage
)
2450 struct object_entry
*trg_entry
= trg
->entry
;
2451 struct object_entry
*src_entry
= src
->entry
;
2452 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
2454 enum object_type type
;
2457 /* Don't bother doing diffs between different types */
2458 if (oe_type(trg_entry
) != oe_type(src_entry
))
2462 * We do not bother to try a delta that we discarded on an
2463 * earlier try, but only when reusing delta data. Note that
2464 * src_entry that is marked as the preferred_base should always
2465 * be considered, as even if we produce a suboptimal delta against
2466 * it, we will still save the transfer cost, as we already know
2467 * the other side has it and we won't send src_entry at all.
2469 if (reuse_delta
&& IN_PACK(trg_entry
) &&
2470 IN_PACK(trg_entry
) == IN_PACK(src_entry
) &&
2471 !src_entry
->preferred_base
&&
2472 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
2473 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
2476 /* Let's not bust the allowed depth. */
2477 if (src
->depth
>= max_depth
)
2480 /* Now some size filtering heuristics. */
2481 trg_size
= SIZE(trg_entry
);
2482 if (!DELTA(trg_entry
)) {
2483 max_size
= trg_size
/2 - the_hash_algo
->rawsz
;
2486 max_size
= DELTA_SIZE(trg_entry
);
2487 ref_depth
= trg
->depth
;
2489 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
2490 (max_depth
- ref_depth
+ 1);
2493 src_size
= SIZE(src_entry
);
2494 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
2495 if (sizediff
>= max_size
)
2497 if (trg_size
< src_size
/ 32)
2500 if (!in_same_island(&trg
->entry
->idx
.oid
, &src
->entry
->idx
.oid
))
2503 /* Load data if not already done */
2505 packing_data_lock(&to_pack
);
2506 trg
->data
= read_object_file(&trg_entry
->idx
.oid
, &type
, &sz
);
2507 packing_data_unlock(&to_pack
);
2509 die(_("object %s cannot be read"),
2510 oid_to_hex(&trg_entry
->idx
.oid
));
2512 die(_("object %s inconsistent object length (%"PRIuMAX
" vs %"PRIuMAX
")"),
2513 oid_to_hex(&trg_entry
->idx
.oid
), (uintmax_t)sz
,
2514 (uintmax_t)trg_size
);
2518 packing_data_lock(&to_pack
);
2519 src
->data
= read_object_file(&src_entry
->idx
.oid
, &type
, &sz
);
2520 packing_data_unlock(&to_pack
);
2522 if (src_entry
->preferred_base
) {
2523 static int warned
= 0;
2525 warning(_("object %s cannot be read"),
2526 oid_to_hex(&src_entry
->idx
.oid
));
2528 * Those objects are not included in the
2529 * resulting pack. Be resilient and ignore
2530 * them if they can't be read, in case the
2531 * pack could be created nevertheless.
2535 die(_("object %s cannot be read"),
2536 oid_to_hex(&src_entry
->idx
.oid
));
2539 die(_("object %s inconsistent object length (%"PRIuMAX
" vs %"PRIuMAX
")"),
2540 oid_to_hex(&src_entry
->idx
.oid
), (uintmax_t)sz
,
2541 (uintmax_t)src_size
);
2545 src
->index
= create_delta_index(src
->data
, src_size
);
2547 static int warned
= 0;
2549 warning(_("suboptimal pack - out of memory"));
2552 *mem_usage
+= sizeof_delta_index(src
->index
);
2555 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
2559 if (DELTA(trg_entry
)) {
2560 /* Prefer only shallower same-sized deltas. */
2561 if (delta_size
== DELTA_SIZE(trg_entry
) &&
2562 src
->depth
+ 1 >= trg
->depth
) {
2569 * Handle memory allocation outside of the cache
2570 * accounting lock. Compiler will optimize the strangeness
2571 * away when NO_PTHREADS is defined.
2573 free(trg_entry
->delta_data
);
2575 if (trg_entry
->delta_data
) {
2576 delta_cache_size
-= DELTA_SIZE(trg_entry
);
2577 trg_entry
->delta_data
= NULL
;
2579 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
2580 delta_cache_size
+= delta_size
;
2582 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
2588 SET_DELTA(trg_entry
, src_entry
);
2589 SET_DELTA_SIZE(trg_entry
, delta_size
);
2590 trg
->depth
= src
->depth
+ 1;
2595 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
2597 struct object_entry
*child
= DELTA_CHILD(me
);
2600 const unsigned int c
= check_delta_limit(child
, n
+ 1);
2603 child
= DELTA_SIBLING(child
);
2608 static unsigned long free_unpacked(struct unpacked
*n
)
2610 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
2611 free_delta_index(n
->index
);
2614 freed_mem
+= SIZE(n
->entry
);
2615 FREE_AND_NULL(n
->data
);
2622 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
2623 int window
, int depth
, unsigned *processed
)
2625 uint32_t i
, idx
= 0, count
= 0;
2626 struct unpacked
*array
;
2627 unsigned long mem_usage
= 0;
2629 CALLOC_ARRAY(array
, window
);
2632 struct object_entry
*entry
;
2633 struct unpacked
*n
= array
+ idx
;
2634 int j
, max_depth
, best_base
= -1;
2643 if (!entry
->preferred_base
) {
2645 display_progress(progress_state
, *processed
);
2649 mem_usage
-= free_unpacked(n
);
2652 while (window_memory_limit
&&
2653 mem_usage
> window_memory_limit
&&
2655 const uint32_t tail
= (idx
+ window
- count
) % window
;
2656 mem_usage
-= free_unpacked(array
+ tail
);
2660 /* We do not compute delta to *create* objects we are not
2663 if (entry
->preferred_base
)
2667 * If the current object is at pack edge, take the depth the
2668 * objects that depend on the current object into account
2669 * otherwise they would become too deep.
2672 if (DELTA_CHILD(entry
)) {
2673 max_depth
-= check_delta_limit(entry
, 0);
2681 uint32_t other_idx
= idx
+ j
;
2683 if (other_idx
>= window
)
2684 other_idx
-= window
;
2685 m
= array
+ other_idx
;
2688 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
2692 best_base
= other_idx
;
2696 * If we decided to cache the delta data, then it is best
2697 * to compress it right away. First because we have to do
2698 * it anyway, and doing it here while we're threaded will
2699 * save a lot of time in the non threaded write phase,
2700 * as well as allow for caching more deltas within
2701 * the same cache size limit.
2703 * But only if not writing to stdout, since in that case
2704 * the network is most likely throttling writes anyway,
2705 * and therefore it is best to go to the write phase ASAP
2706 * instead, as we can afford spending more time compressing
2707 * between writes at that moment.
2709 if (entry
->delta_data
&& !pack_to_stdout
) {
2712 size
= do_compress(&entry
->delta_data
, DELTA_SIZE(entry
));
2713 if (size
< (1U << OE_Z_DELTA_BITS
)) {
2714 entry
->z_delta_size
= size
;
2716 delta_cache_size
-= DELTA_SIZE(entry
);
2717 delta_cache_size
+= entry
->z_delta_size
;
2720 FREE_AND_NULL(entry
->delta_data
);
2721 entry
->z_delta_size
= 0;
2725 /* if we made n a delta, and if n is already at max
2726 * depth, leaving it in the window is pointless. we
2727 * should evict it first.
2729 if (DELTA(entry
) && max_depth
<= n
->depth
)
2733 * Move the best delta base up in the window, after the
2734 * currently deltified object, to keep it longer. It will
2735 * be the first base object to be attempted next.
2738 struct unpacked swap
= array
[best_base
];
2739 int dist
= (window
+ idx
- best_base
) % window
;
2740 int dst
= best_base
;
2742 int src
= (dst
+ 1) % window
;
2743 array
[dst
] = array
[src
];
2751 if (count
+ 1 < window
)
2757 for (i
= 0; i
< window
; ++i
) {
2758 free_delta_index(array
[i
].index
);
2759 free(array
[i
].data
);
2765 * The main object list is split into smaller lists, each is handed to
2768 * The main thread waits on the condition that (at least) one of the workers
2769 * has stopped working (which is indicated in the .working member of
2770 * struct thread_params).
2772 * When a work thread has completed its work, it sets .working to 0 and
2773 * signals the main thread and waits on the condition that .data_ready
2776 * The main thread steals half of the work from the worker that has
2777 * most work left to hand it to the idle worker.
2780 struct thread_params
{
2782 struct object_entry
**list
;
2789 pthread_mutex_t mutex
;
2790 pthread_cond_t cond
;
2791 unsigned *processed
;
2794 static pthread_cond_t progress_cond
;
2797 * Mutex and conditional variable can't be statically-initialized on Windows.
2799 static void init_threaded_search(void)
2801 pthread_mutex_init(&cache_mutex
, NULL
);
2802 pthread_mutex_init(&progress_mutex
, NULL
);
2803 pthread_cond_init(&progress_cond
, NULL
);
2806 static void cleanup_threaded_search(void)
2808 pthread_cond_destroy(&progress_cond
);
2809 pthread_mutex_destroy(&cache_mutex
);
2810 pthread_mutex_destroy(&progress_mutex
);
2813 static void *threaded_find_deltas(void *arg
)
2815 struct thread_params
*me
= arg
;
2818 while (me
->remaining
) {
2821 find_deltas(me
->list
, &me
->remaining
,
2822 me
->window
, me
->depth
, me
->processed
);
2826 pthread_cond_signal(&progress_cond
);
2830 * We must not set ->data_ready before we wait on the
2831 * condition because the main thread may have set it to 1
2832 * before we get here. In order to be sure that new
2833 * work is available if we see 1 in ->data_ready, it
2834 * was initialized to 0 before this thread was spawned
2835 * and we reset it to 0 right away.
2837 pthread_mutex_lock(&me
->mutex
);
2838 while (!me
->data_ready
)
2839 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2841 pthread_mutex_unlock(&me
->mutex
);
2846 /* leave ->working 1 so that this doesn't get more work assigned */
2850 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2851 int window
, int depth
, unsigned *processed
)
2853 struct thread_params
*p
;
2854 int i
, ret
, active_threads
= 0;
2856 init_threaded_search();
2858 if (delta_search_threads
<= 1) {
2859 find_deltas(list
, &list_size
, window
, depth
, processed
);
2860 cleanup_threaded_search();
2863 if (progress
> pack_to_stdout
)
2864 fprintf_ln(stderr
, _("Delta compression using up to %d threads"),
2865 delta_search_threads
);
2866 CALLOC_ARRAY(p
, delta_search_threads
);
2868 /* Partition the work amongst work threads. */
2869 for (i
= 0; i
< delta_search_threads
; i
++) {
2870 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2872 /* don't use too small segments or no deltas will be found */
2873 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2876 p
[i
].window
= window
;
2878 p
[i
].processed
= processed
;
2880 p
[i
].data_ready
= 0;
2882 /* try to split chunks on "path" boundaries */
2883 while (sub_size
&& sub_size
< list_size
&&
2884 list
[sub_size
]->hash
&&
2885 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2889 p
[i
].list_size
= sub_size
;
2890 p
[i
].remaining
= sub_size
;
2893 list_size
-= sub_size
;
2896 /* Start work threads. */
2897 for (i
= 0; i
< delta_search_threads
; i
++) {
2898 if (!p
[i
].list_size
)
2900 pthread_mutex_init(&p
[i
].mutex
, NULL
);
2901 pthread_cond_init(&p
[i
].cond
, NULL
);
2902 ret
= pthread_create(&p
[i
].thread
, NULL
,
2903 threaded_find_deltas
, &p
[i
]);
2905 die(_("unable to create thread: %s"), strerror(ret
));
2910 * Now let's wait for work completion. Each time a thread is done
2911 * with its work, we steal half of the remaining work from the
2912 * thread with the largest number of unprocessed objects and give
2913 * it to that newly idle thread. This ensure good load balancing
2914 * until the remaining object list segments are simply too short
2915 * to be worth splitting anymore.
2917 while (active_threads
) {
2918 struct thread_params
*target
= NULL
;
2919 struct thread_params
*victim
= NULL
;
2920 unsigned sub_size
= 0;
2924 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
2929 pthread_cond_wait(&progress_cond
, &progress_mutex
);
2932 for (i
= 0; i
< delta_search_threads
; i
++)
2933 if (p
[i
].remaining
> 2*window
&&
2934 (!victim
|| victim
->remaining
< p
[i
].remaining
))
2937 sub_size
= victim
->remaining
/ 2;
2938 list
= victim
->list
+ victim
->list_size
- sub_size
;
2939 while (sub_size
&& list
[0]->hash
&&
2940 list
[0]->hash
== list
[-1]->hash
) {
2946 * It is possible for some "paths" to have
2947 * so many objects that no hash boundary
2948 * might be found. Let's just steal the
2949 * exact half in that case.
2951 sub_size
= victim
->remaining
/ 2;
2954 target
->list
= list
;
2955 victim
->list_size
-= sub_size
;
2956 victim
->remaining
-= sub_size
;
2958 target
->list_size
= sub_size
;
2959 target
->remaining
= sub_size
;
2960 target
->working
= 1;
2963 pthread_mutex_lock(&target
->mutex
);
2964 target
->data_ready
= 1;
2965 pthread_cond_signal(&target
->cond
);
2966 pthread_mutex_unlock(&target
->mutex
);
2969 pthread_join(target
->thread
, NULL
);
2970 pthread_cond_destroy(&target
->cond
);
2971 pthread_mutex_destroy(&target
->mutex
);
2975 cleanup_threaded_search();
2979 static int obj_is_packed(const struct object_id
*oid
)
2981 return packlist_find(&to_pack
, oid
) ||
2982 (reuse_packfile_bitmap
&&
2983 bitmap_walk_contains(bitmap_git
, reuse_packfile_bitmap
, oid
));
2986 static void add_tag_chain(const struct object_id
*oid
)
2991 * We catch duplicates already in add_object_entry(), but we'd
2992 * prefer to do this extra check to avoid having to parse the
2993 * tag at all if we already know that it's being packed (e.g., if
2994 * it was included via bitmaps, we would not have parsed it
2997 if (obj_is_packed(oid
))
3000 tag
= lookup_tag(the_repository
, oid
);
3002 if (!tag
|| parse_tag(tag
) || !tag
->tagged
)
3003 die(_("unable to pack objects reachable from tag %s"),
3006 add_object_entry(&tag
->object
.oid
, OBJ_TAG
, NULL
, 0);
3008 if (tag
->tagged
->type
!= OBJ_TAG
)
3011 tag
= (struct tag
*)tag
->tagged
;
3015 static int add_ref_tag(const char *tag
, const struct object_id
*oid
, int flag
, void *cb_data
)
3017 struct object_id peeled
;
3019 if (!peel_iterated_oid(oid
, &peeled
) && obj_is_packed(&peeled
))
3024 static void prepare_pack(int window
, int depth
)
3026 struct object_entry
**delta_list
;
3027 uint32_t i
, nr_deltas
;
3030 if (use_delta_islands
)
3031 resolve_tree_islands(the_repository
, progress
, &to_pack
);
3033 get_object_details();
3036 * If we're locally repacking then we need to be doubly careful
3037 * from now on in order to make sure no stealth corruption gets
3038 * propagated to the new pack. Clients receiving streamed packs
3039 * should validate everything they get anyway so no need to incur
3040 * the additional cost here in that case.
3042 if (!pack_to_stdout
)
3043 do_check_packed_object_crc
= 1;
3045 if (!to_pack
.nr_objects
|| !window
|| !depth
)
3048 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
3051 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
3052 struct object_entry
*entry
= to_pack
.objects
+ i
;
3055 /* This happens if we decided to reuse existing
3056 * delta from a pack. "reuse_delta &&" is implied.
3060 if (!entry
->type_valid
||
3061 oe_size_less_than(&to_pack
, entry
, 50))
3064 if (entry
->no_try_delta
)
3067 if (!entry
->preferred_base
) {
3069 if (oe_type(entry
) < 0)
3070 die(_("unable to get type of object %s"),
3071 oid_to_hex(&entry
->idx
.oid
));
3073 if (oe_type(entry
) < 0) {
3075 * This object is not found, but we
3076 * don't have to include it anyway.
3082 delta_list
[n
++] = entry
;
3085 if (nr_deltas
&& n
> 1) {
3086 unsigned nr_done
= 0;
3089 progress_state
= start_progress(_("Compressing objects"),
3091 QSORT(delta_list
, n
, type_size_sort
);
3092 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
3093 stop_progress(&progress_state
);
3094 if (nr_done
!= nr_deltas
)
3095 die(_("inconsistency with delta count"));
3100 static int git_pack_config(const char *k
, const char *v
, void *cb
)
3102 if (!strcmp(k
, "pack.window")) {
3103 window
= git_config_int(k
, v
);
3106 if (!strcmp(k
, "pack.windowmemory")) {
3107 window_memory_limit
= git_config_ulong(k
, v
);
3110 if (!strcmp(k
, "pack.depth")) {
3111 depth
= git_config_int(k
, v
);
3114 if (!strcmp(k
, "pack.deltacachesize")) {
3115 max_delta_cache_size
= git_config_int(k
, v
);
3118 if (!strcmp(k
, "pack.deltacachelimit")) {
3119 cache_max_small_delta_size
= git_config_int(k
, v
);
3122 if (!strcmp(k
, "pack.writebitmaphashcache")) {
3123 if (git_config_bool(k
, v
))
3124 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
3126 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
3128 if (!strcmp(k
, "pack.usebitmaps")) {
3129 use_bitmap_index_default
= git_config_bool(k
, v
);
3132 if (!strcmp(k
, "pack.allowpackreuse")) {
3133 allow_pack_reuse
= git_config_bool(k
, v
);
3136 if (!strcmp(k
, "pack.threads")) {
3137 delta_search_threads
= git_config_int(k
, v
);
3138 if (delta_search_threads
< 0)
3139 die(_("invalid number of threads specified (%d)"),
3140 delta_search_threads
);
3141 if (!HAVE_THREADS
&& delta_search_threads
!= 1) {
3142 warning(_("no threads support, ignoring %s"), k
);
3143 delta_search_threads
= 0;
3147 if (!strcmp(k
, "pack.indexversion")) {
3148 pack_idx_opts
.version
= git_config_int(k
, v
);
3149 if (pack_idx_opts
.version
> 2)
3150 die(_("bad pack.indexversion=%"PRIu32
),
3151 pack_idx_opts
.version
);
3154 if (!strcmp(k
, "pack.writereverseindex")) {
3155 if (git_config_bool(k
, v
))
3156 pack_idx_opts
.flags
|= WRITE_REV
;
3158 pack_idx_opts
.flags
&= ~WRITE_REV
;
3161 if (!strcmp(k
, "uploadpack.blobpackfileuri")) {
3162 struct configured_exclusion
*ex
= xmalloc(sizeof(*ex
));
3163 const char *oid_end
, *pack_end
;
3165 * Stores the pack hash. This is not a true object ID, but is
3168 struct object_id pack_hash
;
3170 if (parse_oid_hex(v
, &ex
->e
.oid
, &oid_end
) ||
3172 parse_oid_hex(oid_end
+ 1, &pack_hash
, &pack_end
) ||
3174 die(_("value of uploadpack.blobpackfileuri must be "
3175 "of the form '<object-hash> <pack-hash> <uri>' (got '%s')"), v
);
3176 if (oidmap_get(&configured_exclusions
, &ex
->e
.oid
))
3177 die(_("object already configured in another "
3178 "uploadpack.blobpackfileuri (got '%s')"), v
);
3179 ex
->pack_hash_hex
= xcalloc(1, pack_end
- oid_end
);
3180 memcpy(ex
->pack_hash_hex
, oid_end
+ 1, pack_end
- oid_end
- 1);
3181 ex
->uri
= xstrdup(pack_end
+ 1);
3182 oidmap_put(&configured_exclusions
, ex
);
3184 return git_default_config(k
, v
, cb
);
3187 /* Counters for trace2 output when in --stdin-packs mode. */
3188 static int stdin_packs_found_nr
;
3189 static int stdin_packs_hints_nr
;
3191 static int add_object_entry_from_pack(const struct object_id
*oid
,
3192 struct packed_git
*p
,
3196 struct rev_info
*revs
= _data
;
3197 struct object_info oi
= OBJECT_INFO_INIT
;
3199 enum object_type type
;
3201 display_progress(progress_state
, ++nr_seen
);
3203 if (have_duplicate_entry(oid
, 0))
3206 ofs
= nth_packed_object_offset(p
, pos
);
3207 if (!want_object_in_pack(oid
, 0, &p
, &ofs
))
3211 if (packed_object_info(the_repository
, p
, ofs
, &oi
) < 0)
3212 die(_("could not get type of object %s in pack %s"),
3213 oid_to_hex(oid
), p
->pack_name
);
3214 else if (type
== OBJ_COMMIT
) {
3216 * commits in included packs are used as starting points for the
3217 * subsequent revision walk
3219 add_pending_oid(revs
, NULL
, oid
, 0);
3222 stdin_packs_found_nr
++;
3224 create_object_entry(oid
, type
, 0, 0, 0, p
, ofs
);
3229 static void show_commit_pack_hint(struct commit
*commit
, void *_data
)
3231 /* nothing to do; commits don't have a namehash */
3234 static void show_object_pack_hint(struct object
*object
, const char *name
,
3237 struct object_entry
*oe
= packlist_find(&to_pack
, &object
->oid
);
3242 * Our 'to_pack' list was constructed by iterating all objects packed in
3243 * included packs, and so doesn't have a non-zero hash field that you
3244 * would typically pick up during a reachability traversal.
3246 * Make a best-effort attempt to fill in the ->hash and ->no_try_delta
3247 * here using a now in order to perhaps improve the delta selection
3250 oe
->hash
= pack_name_hash(name
);
3251 oe
->no_try_delta
= name
&& no_try_delta(name
);
3253 stdin_packs_hints_nr
++;
3256 static int pack_mtime_cmp(const void *_a
, const void *_b
)
3258 struct packed_git
*a
= ((const struct string_list_item
*)_a
)->util
;
3259 struct packed_git
*b
= ((const struct string_list_item
*)_b
)->util
;
3262 * order packs by descending mtime so that objects are laid out
3263 * roughly as newest-to-oldest
3265 if (a
->mtime
< b
->mtime
)
3267 else if (b
->mtime
< a
->mtime
)
3273 static void read_packs_list_from_stdin(void)
3275 struct strbuf buf
= STRBUF_INIT
;
3276 struct string_list include_packs
= STRING_LIST_INIT_DUP
;
3277 struct string_list exclude_packs
= STRING_LIST_INIT_DUP
;
3278 struct string_list_item
*item
= NULL
;
3280 struct packed_git
*p
;
3281 struct rev_info revs
;
3283 repo_init_revisions(the_repository
, &revs
, NULL
);
3285 * Use a revision walk to fill in the namehash of objects in the include
3286 * packs. To save time, we'll avoid traversing through objects that are
3287 * in excluded packs.
3289 * That may cause us to avoid populating all of the namehash fields of
3290 * all included objects, but our goal is best-effort, since this is only
3291 * an optimization during delta selection.
3293 revs
.no_kept_objects
= 1;
3294 revs
.keep_pack_cache_flags
|= IN_CORE_KEEP_PACKS
;
3295 revs
.blob_objects
= 1;
3296 revs
.tree_objects
= 1;
3297 revs
.tag_objects
= 1;
3298 revs
.ignore_missing_links
= 1;
3300 while (strbuf_getline(&buf
, stdin
) != EOF
) {
3304 if (*buf
.buf
== '^')
3305 string_list_append(&exclude_packs
, buf
.buf
+ 1);
3307 string_list_append(&include_packs
, buf
.buf
);
3312 string_list_sort(&include_packs
);
3313 string_list_sort(&exclude_packs
);
3315 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3316 const char *pack_name
= pack_basename(p
);
3318 item
= string_list_lookup(&include_packs
, pack_name
);
3320 item
= string_list_lookup(&exclude_packs
, pack_name
);
3327 * Arguments we got on stdin may not even be packs. First
3328 * check that to avoid segfaulting later on in
3329 * e.g. pack_mtime_cmp(), excluded packs are handled below.
3331 * Since we first parsed our STDIN and then sorted the input
3332 * lines the pack we error on will be whatever line happens to
3333 * sort first. This is lazy, it's enough that we report one
3334 * bad case here, we don't need to report the first/last one,
3337 for_each_string_list_item(item
, &include_packs
) {
3338 struct packed_git
*p
= item
->util
;
3340 die(_("could not find pack '%s'"), item
->string
);
3344 * Then, handle all of the excluded packs, marking them as
3345 * kept in-core so that later calls to add_object_entry()
3346 * discards any objects that are also found in excluded packs.
3348 for_each_string_list_item(item
, &exclude_packs
) {
3349 struct packed_git
*p
= item
->util
;
3351 die(_("could not find pack '%s'"), item
->string
);
3352 p
->pack_keep_in_core
= 1;
3356 * Order packs by ascending mtime; use QSORT directly to access the
3357 * string_list_item's ->util pointer, which string_list_sort() does not
3360 QSORT(include_packs
.items
, include_packs
.nr
, pack_mtime_cmp
);
3362 for_each_string_list_item(item
, &include_packs
) {
3363 struct packed_git
*p
= item
->util
;
3365 die(_("could not find pack '%s'"), item
->string
);
3366 for_each_object_in_pack(p
,
3367 add_object_entry_from_pack
,
3369 FOR_EACH_OBJECT_PACK_ORDER
);
3372 if (prepare_revision_walk(&revs
))
3373 die(_("revision walk setup failed"));
3374 traverse_commit_list(&revs
,
3375 show_commit_pack_hint
,
3376 show_object_pack_hint
,
3379 trace2_data_intmax("pack-objects", the_repository
, "stdin_packs_found",
3380 stdin_packs_found_nr
);
3381 trace2_data_intmax("pack-objects", the_repository
, "stdin_packs_hints",
3382 stdin_packs_hints_nr
);
3384 strbuf_release(&buf
);
3385 string_list_clear(&include_packs
, 0);
3386 string_list_clear(&exclude_packs
, 0);
3389 static void read_object_list_from_stdin(void)
3391 char line
[GIT_MAX_HEXSZ
+ 1 + PATH_MAX
+ 2];
3392 struct object_id oid
;
3396 if (!fgets(line
, sizeof(line
), stdin
)) {
3400 BUG("fgets returned NULL, not EOF, not error!");
3406 if (line
[0] == '-') {
3407 if (get_oid_hex(line
+1, &oid
))
3408 die(_("expected edge object ID, got garbage:\n %s"),
3410 add_preferred_base(&oid
);
3413 if (parse_oid_hex(line
, &oid
, &p
))
3414 die(_("expected object ID, got garbage:\n %s"), line
);
3416 add_preferred_base_object(p
+ 1);
3417 add_object_entry(&oid
, OBJ_NONE
, p
+ 1, 0);
3421 static void show_commit(struct commit
*commit
, void *data
)
3423 add_object_entry(&commit
->object
.oid
, OBJ_COMMIT
, NULL
, 0);
3425 if (write_bitmap_index
)
3426 index_commit_for_bitmap(commit
);
3428 if (use_delta_islands
)
3429 propagate_island_marks(commit
);
3432 static void show_object(struct object
*obj
, const char *name
, void *data
)
3434 add_preferred_base_object(name
);
3435 add_object_entry(&obj
->oid
, obj
->type
, name
, 0);
3437 if (use_delta_islands
) {
3440 struct object_entry
*ent
;
3442 /* the empty string is a root tree, which is depth 0 */
3443 depth
= *name
? 1 : 0;
3444 for (p
= strchr(name
, '/'); p
; p
= strchr(p
+ 1, '/'))
3447 ent
= packlist_find(&to_pack
, &obj
->oid
);
3448 if (ent
&& depth
> oe_tree_depth(&to_pack
, ent
))
3449 oe_set_tree_depth(&to_pack
, ent
, depth
);
3453 static void show_object__ma_allow_any(struct object
*obj
, const char *name
, void *data
)
3455 assert(arg_missing_action
== MA_ALLOW_ANY
);
3458 * Quietly ignore ALL missing objects. This avoids problems with
3459 * staging them now and getting an odd error later.
3461 if (!has_object(the_repository
, &obj
->oid
, 0))
3464 show_object(obj
, name
, data
);
3467 static void show_object__ma_allow_promisor(struct object
*obj
, const char *name
, void *data
)
3469 assert(arg_missing_action
== MA_ALLOW_PROMISOR
);
3472 * Quietly ignore EXPECTED missing objects. This avoids problems with
3473 * staging them now and getting an odd error later.
3475 if (!has_object(the_repository
, &obj
->oid
, 0) && is_promisor_object(&obj
->oid
))
3478 show_object(obj
, name
, data
);
3481 static int option_parse_missing_action(const struct option
*opt
,
3482 const char *arg
, int unset
)
3487 if (!strcmp(arg
, "error")) {
3488 arg_missing_action
= MA_ERROR
;
3489 fn_show_object
= show_object
;
3493 if (!strcmp(arg
, "allow-any")) {
3494 arg_missing_action
= MA_ALLOW_ANY
;
3495 fetch_if_missing
= 0;
3496 fn_show_object
= show_object__ma_allow_any
;
3500 if (!strcmp(arg
, "allow-promisor")) {
3501 arg_missing_action
= MA_ALLOW_PROMISOR
;
3502 fetch_if_missing
= 0;
3503 fn_show_object
= show_object__ma_allow_promisor
;
3507 die(_("invalid value for --missing"));
3511 static void show_edge(struct commit
*commit
)
3513 add_preferred_base(&commit
->object
.oid
);
3516 static int add_object_in_unpacked_pack(const struct object_id
*oid
,
3517 struct packed_git
*pack
,
3521 add_object_entry(oid
, OBJ_NONE
, "", 0);
3525 static void add_objects_in_unpacked_packs(void)
3527 if (for_each_packed_object(add_object_in_unpacked_pack
, NULL
,
3528 FOR_EACH_OBJECT_PACK_ORDER
|
3529 FOR_EACH_OBJECT_LOCAL_ONLY
|
3530 FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS
|
3531 FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS
))
3532 die(_("cannot open pack index"));
3535 static int add_loose_object(const struct object_id
*oid
, const char *path
,
3538 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
3541 warning(_("loose object at %s could not be examined"), path
);
3545 add_object_entry(oid
, type
, "", 0);
3550 * We actually don't even have to worry about reachability here.
3551 * add_object_entry will weed out duplicates, so we just add every
3552 * loose object we find.
3554 static void add_unreachable_loose_objects(void)
3556 for_each_loose_file_in_objdir(get_object_directory(),
3561 static int has_sha1_pack_kept_or_nonlocal(const struct object_id
*oid
)
3563 static struct packed_git
*last_found
= (void *)1;
3564 struct packed_git
*p
;
3566 p
= (last_found
!= (void *)1) ? last_found
:
3567 get_all_packs(the_repository
);
3570 if ((!p
->pack_local
|| p
->pack_keep
||
3571 p
->pack_keep_in_core
) &&
3572 find_pack_entry_one(oid
->hash
, p
)) {
3576 if (p
== last_found
)
3577 p
= get_all_packs(the_repository
);
3580 if (p
== last_found
)
3587 * Store a list of sha1s that are should not be discarded
3588 * because they are either written too recently, or are
3589 * reachable from another object that was.
3591 * This is filled by get_object_list.
3593 static struct oid_array recent_objects
;
3595 static int loosened_object_can_be_discarded(const struct object_id
*oid
,
3598 if (!unpack_unreachable_expiration
)
3600 if (mtime
> unpack_unreachable_expiration
)
3602 if (oid_array_lookup(&recent_objects
, oid
) >= 0)
3607 static void loosen_unused_packed_objects(void)
3609 struct packed_git
*p
;
3611 uint32_t loosened_objects_nr
= 0;
3612 struct object_id oid
;
3614 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3615 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
3618 if (open_pack_index(p
))
3619 die(_("cannot open pack index"));
3621 for (i
= 0; i
< p
->num_objects
; i
++) {
3622 nth_packed_object_id(&oid
, p
, i
);
3623 if (!packlist_find(&to_pack
, &oid
) &&
3624 !has_sha1_pack_kept_or_nonlocal(&oid
) &&
3625 !loosened_object_can_be_discarded(&oid
, p
->mtime
)) {
3626 if (force_object_loose(&oid
, p
->mtime
))
3627 die(_("unable to force loose object"));
3628 loosened_objects_nr
++;
3633 trace2_data_intmax("pack-objects", the_repository
,
3634 "loosen_unused_packed_objects/loosened", loosened_objects_nr
);
3638 * This tracks any options which pack-reuse code expects to be on, or which a
3639 * reader of the pack might not understand, and which would therefore prevent
3640 * blind reuse of what we have on disk.
3642 static int pack_options_allow_reuse(void)
3644 return allow_pack_reuse
&&
3646 !ignore_packed_keep_on_disk
&&
3647 !ignore_packed_keep_in_core
&&
3648 (!local
|| !have_non_local_packs
) &&
3652 static int get_object_list_from_bitmap(struct rev_info
*revs
)
3654 if (!(bitmap_git
= prepare_bitmap_walk(revs
, &filter_options
, 0)))
3657 if (pack_options_allow_reuse() &&
3658 !reuse_partial_packfile_from_bitmap(
3661 &reuse_packfile_objects
,
3662 &reuse_packfile_bitmap
)) {
3663 assert(reuse_packfile_objects
);
3664 nr_result
+= reuse_packfile_objects
;
3665 nr_seen
+= reuse_packfile_objects
;
3666 display_progress(progress_state
, nr_seen
);
3669 traverse_bitmap_commit_list(bitmap_git
, revs
,
3670 &add_object_entry_from_bitmap
);
3674 static void record_recent_object(struct object
*obj
,
3678 oid_array_append(&recent_objects
, &obj
->oid
);
3681 static void record_recent_commit(struct commit
*commit
, void *data
)
3683 oid_array_append(&recent_objects
, &commit
->object
.oid
);
3686 static int mark_bitmap_preferred_tip(const char *refname
,
3687 const struct object_id
*oid
, int flags
,
3690 struct object_id peeled
;
3691 struct object
*object
;
3693 if (!peel_iterated_oid(oid
, &peeled
))
3696 object
= parse_object_or_die(oid
, refname
);
3697 if (object
->type
== OBJ_COMMIT
)
3698 object
->flags
|= NEEDS_BITMAP
;
3703 static void mark_bitmap_preferred_tips(void)
3705 struct string_list_item
*item
;
3706 const struct string_list
*preferred_tips
;
3708 preferred_tips
= bitmap_preferred_tips(the_repository
);
3709 if (!preferred_tips
)
3712 for_each_string_list_item(item
, preferred_tips
) {
3713 for_each_ref_in(item
->string
, mark_bitmap_preferred_tip
, NULL
);
3717 static void get_object_list(int ac
, const char **av
)
3719 struct rev_info revs
;
3720 struct setup_revision_opt s_r_opt
= {
3721 .allow_exclude_promisor_objects
= 1,
3727 repo_init_revisions(the_repository
, &revs
, NULL
);
3728 save_commit_buffer
= 0;
3729 setup_revisions(ac
, av
, &revs
, &s_r_opt
);
3731 /* make sure shallows are read */
3732 is_repository_shallow(the_repository
);
3734 save_warning
= warn_on_object_refname_ambiguity
;
3735 warn_on_object_refname_ambiguity
= 0;
3737 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
3738 int len
= strlen(line
);
3739 if (len
&& line
[len
- 1] == '\n')
3744 if (!strcmp(line
, "--not")) {
3745 flags
^= UNINTERESTING
;
3746 write_bitmap_index
= 0;
3749 if (starts_with(line
, "--shallow ")) {
3750 struct object_id oid
;
3751 if (get_oid_hex(line
+ 10, &oid
))
3752 die("not an object name '%s'", line
+ 10);
3753 register_shallow(the_repository
, &oid
);
3754 use_bitmap_index
= 0;
3757 die(_("not a rev '%s'"), line
);
3759 if (handle_revision_arg(line
, &revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
3760 die(_("bad revision '%s'"), line
);
3763 warn_on_object_refname_ambiguity
= save_warning
;
3765 if (use_bitmap_index
&& !get_object_list_from_bitmap(&revs
))
3768 if (use_delta_islands
)
3769 load_delta_islands(the_repository
, progress
);
3771 if (write_bitmap_index
)
3772 mark_bitmap_preferred_tips();
3774 if (prepare_revision_walk(&revs
))
3775 die(_("revision walk setup failed"));
3776 mark_edges_uninteresting(&revs
, show_edge
, sparse
);
3778 if (!fn_show_object
)
3779 fn_show_object
= show_object
;
3780 traverse_commit_list_filtered(&filter_options
, &revs
,
3781 show_commit
, fn_show_object
, NULL
,
3784 if (unpack_unreachable_expiration
) {
3785 revs
.ignore_missing_links
= 1;
3786 if (add_unseen_recent_objects_to_traversal(&revs
,
3787 unpack_unreachable_expiration
))
3788 die(_("unable to add recent objects"));
3789 if (prepare_revision_walk(&revs
))
3790 die(_("revision walk setup failed"));
3791 traverse_commit_list(&revs
, record_recent_commit
,
3792 record_recent_object
, NULL
);
3795 if (keep_unreachable
)
3796 add_objects_in_unpacked_packs();
3797 if (pack_loose_unreachable
)
3798 add_unreachable_loose_objects();
3799 if (unpack_unreachable
)
3800 loosen_unused_packed_objects();
3802 oid_array_clear(&recent_objects
);
3805 static void add_extra_kept_packs(const struct string_list
*names
)
3807 struct packed_git
*p
;
3812 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3813 const char *name
= basename(p
->pack_name
);
3819 for (i
= 0; i
< names
->nr
; i
++)
3820 if (!fspathcmp(name
, names
->items
[i
].string
))
3823 if (i
< names
->nr
) {
3824 p
->pack_keep_in_core
= 1;
3825 ignore_packed_keep_in_core
= 1;
3831 static int option_parse_index_version(const struct option
*opt
,
3832 const char *arg
, int unset
)
3835 const char *val
= arg
;
3837 BUG_ON_OPT_NEG(unset
);
3839 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
3840 if (pack_idx_opts
.version
> 2)
3841 die(_("unsupported index version %s"), val
);
3842 if (*c
== ',' && c
[1])
3843 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
3844 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
3845 die(_("bad index version '%s'"), val
);
3849 static int option_parse_unpack_unreachable(const struct option
*opt
,
3850 const char *arg
, int unset
)
3853 unpack_unreachable
= 0;
3854 unpack_unreachable_expiration
= 0;
3857 unpack_unreachable
= 1;
3859 unpack_unreachable_expiration
= approxidate(arg
);
3864 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
3866 int use_internal_rev_list
= 0;
3868 int all_progress_implied
= 0;
3869 struct strvec rp
= STRVEC_INIT
;
3870 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
3871 int rev_list_index
= 0;
3872 int stdin_packs
= 0;
3873 struct string_list keep_pack_list
= STRING_LIST_INIT_NODUP
;
3874 struct option pack_objects_options
[] = {
3875 OPT_SET_INT('q', "quiet", &progress
,
3876 N_("do not show progress meter"), 0),
3877 OPT_SET_INT(0, "progress", &progress
,
3878 N_("show progress meter"), 1),
3879 OPT_SET_INT(0, "all-progress", &progress
,
3880 N_("show progress meter during object writing phase"), 2),
3881 OPT_BOOL(0, "all-progress-implied",
3882 &all_progress_implied
,
3883 N_("similar to --all-progress when progress meter is shown")),
3884 OPT_CALLBACK_F(0, "index-version", NULL
, N_("<version>[,<offset>]"),
3885 N_("write the pack index file in the specified idx format version"),
3886 PARSE_OPT_NONEG
, option_parse_index_version
),
3887 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
3888 N_("maximum size of each output pack file")),
3889 OPT_BOOL(0, "local", &local
,
3890 N_("ignore borrowed objects from alternate object store")),
3891 OPT_BOOL(0, "incremental", &incremental
,
3892 N_("ignore packed objects")),
3893 OPT_INTEGER(0, "window", &window
,
3894 N_("limit pack window by objects")),
3895 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
3896 N_("limit pack window by memory in addition to object limit")),
3897 OPT_INTEGER(0, "depth", &depth
,
3898 N_("maximum length of delta chain allowed in the resulting pack")),
3899 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
3900 N_("reuse existing deltas")),
3901 OPT_BOOL(0, "reuse-object", &reuse_object
,
3902 N_("reuse existing objects")),
3903 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
3904 N_("use OFS_DELTA objects")),
3905 OPT_INTEGER(0, "threads", &delta_search_threads
,
3906 N_("use threads when searching for best delta matches")),
3907 OPT_BOOL(0, "non-empty", &non_empty
,
3908 N_("do not create an empty pack output")),
3909 OPT_BOOL(0, "revs", &use_internal_rev_list
,
3910 N_("read revision arguments from standard input")),
3911 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked
,
3912 N_("limit the objects to those that are not yet packed"),
3913 1, PARSE_OPT_NONEG
),
3914 OPT_SET_INT_F(0, "all", &rev_list_all
,
3915 N_("include objects reachable from any reference"),
3916 1, PARSE_OPT_NONEG
),
3917 OPT_SET_INT_F(0, "reflog", &rev_list_reflog
,
3918 N_("include objects referred by reflog entries"),
3919 1, PARSE_OPT_NONEG
),
3920 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index
,
3921 N_("include objects referred to by the index"),
3922 1, PARSE_OPT_NONEG
),
3923 OPT_BOOL(0, "stdin-packs", &stdin_packs
,
3924 N_("read packs from stdin")),
3925 OPT_BOOL(0, "stdout", &pack_to_stdout
,
3926 N_("output pack to stdout")),
3927 OPT_BOOL(0, "include-tag", &include_tag
,
3928 N_("include tag objects that refer to objects to be packed")),
3929 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
3930 N_("keep unreachable objects")),
3931 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
3932 N_("pack loose unreachable objects")),
3933 OPT_CALLBACK_F(0, "unpack-unreachable", NULL
, N_("time"),
3934 N_("unpack unreachable objects newer than <time>"),
3935 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
),
3936 OPT_BOOL(0, "sparse", &sparse
,
3937 N_("use the sparse reachability algorithm")),
3938 OPT_BOOL(0, "thin", &thin
,
3939 N_("create thin packs")),
3940 OPT_BOOL(0, "shallow", &shallow
,
3941 N_("create packs suitable for shallow fetches")),
3942 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk
,
3943 N_("ignore packs that have companion .keep file")),
3944 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list
, N_("name"),
3945 N_("ignore this pack")),
3946 OPT_INTEGER(0, "compression", &pack_compression_level
,
3947 N_("pack compression level")),
3948 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
3949 N_("do not hide commits by grafts"), 0),
3950 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
3951 N_("use a bitmap index if available to speed up counting objects")),
3952 OPT_SET_INT(0, "write-bitmap-index", &write_bitmap_index
,
3953 N_("write a bitmap index together with the pack index"),
3955 OPT_SET_INT_F(0, "write-bitmap-index-quiet",
3956 &write_bitmap_index
,
3957 N_("write a bitmap index if possible"),
3958 WRITE_BITMAP_QUIET
, PARSE_OPT_HIDDEN
),
3959 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options
),
3960 OPT_CALLBACK_F(0, "missing", NULL
, N_("action"),
3961 N_("handling for missing objects"), PARSE_OPT_NONEG
,
3962 option_parse_missing_action
),
3963 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects
,
3964 N_("do not pack objects in promisor packfiles")),
3965 OPT_BOOL(0, "delta-islands", &use_delta_islands
,
3966 N_("respect islands during delta compression")),
3967 OPT_STRING_LIST(0, "uri-protocol", &uri_protocols
,
3969 N_("exclude any configured uploadpack.blobpackfileuri with this protocol")),
3973 if (DFS_NUM_STATES
> (1 << OE_DFS_STATE_BITS
))
3974 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3976 read_replace_refs
= 0;
3978 sparse
= git_env_bool("GIT_TEST_PACK_SPARSE", -1);
3979 if (the_repository
->gitdir
) {
3980 prepare_repo_settings(the_repository
);
3982 sparse
= the_repository
->settings
.pack_use_sparse
;
3985 reset_pack_idx_option(&pack_idx_opts
);
3986 git_config(git_pack_config
, NULL
);
3987 if (git_env_bool(GIT_TEST_WRITE_REV_INDEX
, 0))
3988 pack_idx_opts
.flags
|= WRITE_REV
;
3990 progress
= isatty(2);
3991 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
3995 base_name
= argv
[0];
3998 if (pack_to_stdout
!= !base_name
|| argc
)
3999 usage_with_options(pack_usage
, pack_objects_options
);
4003 if (depth
>= (1 << OE_DEPTH_BITS
)) {
4004 warning(_("delta chain depth %d is too deep, forcing %d"),
4005 depth
, (1 << OE_DEPTH_BITS
) - 1);
4006 depth
= (1 << OE_DEPTH_BITS
) - 1;
4008 if (cache_max_small_delta_size
>= (1U << OE_Z_DELTA_BITS
)) {
4009 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
4010 (1U << OE_Z_DELTA_BITS
) - 1);
4011 cache_max_small_delta_size
= (1U << OE_Z_DELTA_BITS
) - 1;
4016 strvec_push(&rp
, "pack-objects");
4018 use_internal_rev_list
= 1;
4019 strvec_push(&rp
, shallow
4020 ? "--objects-edge-aggressive"
4021 : "--objects-edge");
4023 strvec_push(&rp
, "--objects");
4026 use_internal_rev_list
= 1;
4027 strvec_push(&rp
, "--all");
4029 if (rev_list_reflog
) {
4030 use_internal_rev_list
= 1;
4031 strvec_push(&rp
, "--reflog");
4033 if (rev_list_index
) {
4034 use_internal_rev_list
= 1;
4035 strvec_push(&rp
, "--indexed-objects");
4037 if (rev_list_unpacked
&& !stdin_packs
) {
4038 use_internal_rev_list
= 1;
4039 strvec_push(&rp
, "--unpacked");
4042 if (exclude_promisor_objects
) {
4043 use_internal_rev_list
= 1;
4044 fetch_if_missing
= 0;
4045 strvec_push(&rp
, "--exclude-promisor-objects");
4047 if (unpack_unreachable
|| keep_unreachable
|| pack_loose_unreachable
)
4048 use_internal_rev_list
= 1;
4052 if (pack_compression_level
== -1)
4053 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
4054 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
4055 die(_("bad pack compression level %d"), pack_compression_level
);
4057 if (!delta_search_threads
) /* --threads=0 means autodetect */
4058 delta_search_threads
= online_cpus();
4060 if (!HAVE_THREADS
&& delta_search_threads
!= 1)
4061 warning(_("no threads support, ignoring --threads"));
4062 if (!pack_to_stdout
&& !pack_size_limit
)
4063 pack_size_limit
= pack_size_limit_cfg
;
4064 if (pack_to_stdout
&& pack_size_limit
)
4065 die(_("--max-pack-size cannot be used to build a pack for transfer"));
4066 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
4067 warning(_("minimum pack size limit is 1 MiB"));
4068 pack_size_limit
= 1024*1024;
4071 if (!pack_to_stdout
&& thin
)
4072 die(_("--thin cannot be used to build an indexable pack"));
4074 if (keep_unreachable
&& unpack_unreachable
)
4075 die(_("options '%s' and '%s' cannot be used together"), "--keep-unreachable", "--unpack-unreachable");
4076 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
4077 unpack_unreachable_expiration
= 0;
4079 if (filter_options
.choice
) {
4080 if (!pack_to_stdout
)
4081 die(_("cannot use --filter without --stdout"));
4083 die(_("cannot use --filter with --stdin-packs"));
4086 if (stdin_packs
&& use_internal_rev_list
)
4087 die(_("cannot use internal rev list with --stdin-packs"));
4090 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
4092 * - to produce good pack (with bitmap index not-yet-packed objects are
4093 * packed in suboptimal order).
4095 * - to use more robust pack-generation codepath (avoiding possible
4096 * bugs in bitmap code and possible bitmap index corruption).
4098 if (!pack_to_stdout
)
4099 use_bitmap_index_default
= 0;
4101 if (use_bitmap_index
< 0)
4102 use_bitmap_index
= use_bitmap_index_default
;
4104 /* "hard" reasons not to use bitmaps; these just won't work at all */
4105 if (!use_internal_rev_list
|| (!pack_to_stdout
&& write_bitmap_index
) || is_repository_shallow(the_repository
))
4106 use_bitmap_index
= 0;
4108 if (pack_to_stdout
|| !rev_list_all
)
4109 write_bitmap_index
= 0;
4111 if (use_delta_islands
)
4112 strvec_push(&rp
, "--topo-order");
4114 if (progress
&& all_progress_implied
)
4117 add_extra_kept_packs(&keep_pack_list
);
4118 if (ignore_packed_keep_on_disk
) {
4119 struct packed_git
*p
;
4120 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
)
4121 if (p
->pack_local
&& p
->pack_keep
)
4123 if (!p
) /* no keep-able packs found */
4124 ignore_packed_keep_on_disk
= 0;
4128 * unlike ignore_packed_keep_on_disk above, we do not
4129 * want to unset "local" based on looking at packs, as
4130 * it also covers non-local objects
4132 struct packed_git
*p
;
4133 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
4134 if (!p
->pack_local
) {
4135 have_non_local_packs
= 1;
4141 trace2_region_enter("pack-objects", "enumerate-objects",
4143 prepare_packing_data(the_repository
, &to_pack
);
4146 progress_state
= start_progress(_("Enumerating objects"), 0);
4148 /* avoids adding objects in excluded packs */
4149 ignore_packed_keep_in_core
= 1;
4150 read_packs_list_from_stdin();
4151 if (rev_list_unpacked
)
4152 add_unreachable_loose_objects();
4153 } else if (!use_internal_rev_list
) {
4154 read_object_list_from_stdin();
4156 get_object_list(rp
.nr
, rp
.v
);
4158 cleanup_preferred_base();
4159 if (include_tag
&& nr_result
)
4160 for_each_tag_ref(add_ref_tag
, NULL
);
4161 stop_progress(&progress_state
);
4162 trace2_region_leave("pack-objects", "enumerate-objects",
4165 if (non_empty
&& !nr_result
)
4168 trace2_region_enter("pack-objects", "prepare-pack",
4170 prepare_pack(window
, depth
);
4171 trace2_region_leave("pack-objects", "prepare-pack",
4175 trace2_region_enter("pack-objects", "write-pack-file", the_repository
);
4176 write_excluded_by_configs();
4178 trace2_region_leave("pack-objects", "write-pack-file", the_repository
);
4182 _("Total %"PRIu32
" (delta %"PRIu32
"),"
4183 " reused %"PRIu32
" (delta %"PRIu32
"),"
4184 " pack-reused %"PRIu32
),
4185 written
, written_delta
, reused
, reused_delta
,
4186 reuse_packfile_objects
);