2 #include "environment.h"
5 #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-options.h"
20 #include "pack-objects.h"
23 #include "streaming.h"
24 #include "thread-utils.h"
25 #include "pack-bitmap.h"
26 #include "delta-islands.h"
27 #include "reachable.h"
28 #include "oid-array.h"
32 #include "object-file.h"
33 #include "object-store-ll.h"
34 #include "replace-object.h"
39 #include "promisor-remote.h"
40 #include "pack-mtimes.h"
41 #include "parse-options.h"
44 * Objects we are going to pack are collected in the `to_pack` structure.
45 * It contains an array (dynamically expanded) of the object data, and a map
46 * that can resolve SHA1s to their position in the array.
48 static struct packing_data to_pack
;
50 static inline struct object_entry
*oe_delta(
51 const struct packing_data
*pack
,
52 const struct object_entry
*e
)
57 return &pack
->ext_bases
[e
->delta_idx
- 1];
59 return &pack
->objects
[e
->delta_idx
- 1];
62 static inline unsigned long oe_delta_size(struct packing_data
*pack
,
63 const struct object_entry
*e
)
65 if (e
->delta_size_valid
)
66 return e
->delta_size_
;
69 * pack->delta_size[] can't be NULL because oe_set_delta_size()
70 * must have been called when a new delta is saved with
72 * If oe_delta() returns NULL (i.e. default state, which means
73 * delta_size_valid is also false), then the caller must never
74 * call oe_delta_size().
76 return pack
->delta_size
[e
- pack
->objects
];
79 unsigned long oe_get_size_slow(struct packing_data
*pack
,
80 const struct object_entry
*e
);
82 static inline unsigned long oe_size(struct packing_data
*pack
,
83 const struct object_entry
*e
)
88 return oe_get_size_slow(pack
, e
);
91 static inline void oe_set_delta(struct packing_data
*pack
,
92 struct object_entry
*e
,
93 struct object_entry
*delta
)
96 e
->delta_idx
= (delta
- pack
->objects
) + 1;
101 static inline struct object_entry
*oe_delta_sibling(
102 const struct packing_data
*pack
,
103 const struct object_entry
*e
)
105 if (e
->delta_sibling_idx
)
106 return &pack
->objects
[e
->delta_sibling_idx
- 1];
110 static inline struct object_entry
*oe_delta_child(
111 const struct packing_data
*pack
,
112 const struct object_entry
*e
)
114 if (e
->delta_child_idx
)
115 return &pack
->objects
[e
->delta_child_idx
- 1];
119 static inline void oe_set_delta_child(struct packing_data
*pack
,
120 struct object_entry
*e
,
121 struct object_entry
*delta
)
124 e
->delta_child_idx
= (delta
- pack
->objects
) + 1;
126 e
->delta_child_idx
= 0;
129 static inline void oe_set_delta_sibling(struct packing_data
*pack
,
130 struct object_entry
*e
,
131 struct object_entry
*delta
)
134 e
->delta_sibling_idx
= (delta
- pack
->objects
) + 1;
136 e
->delta_sibling_idx
= 0;
139 static inline void oe_set_size(struct packing_data
*pack
,
140 struct object_entry
*e
,
143 if (size
< pack
->oe_size_limit
) {
148 if (oe_get_size_slow(pack
, e
) != size
)
149 BUG("'size' is supposed to be the object size!");
153 static inline void oe_set_delta_size(struct packing_data
*pack
,
154 struct object_entry
*e
,
157 if (size
< pack
->oe_delta_size_limit
) {
158 e
->delta_size_
= size
;
159 e
->delta_size_valid
= 1;
161 packing_data_lock(pack
);
162 if (!pack
->delta_size
)
163 ALLOC_ARRAY(pack
->delta_size
, pack
->nr_alloc
);
164 packing_data_unlock(pack
);
166 pack
->delta_size
[e
- pack
->objects
] = size
;
167 e
->delta_size_valid
= 0;
171 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
172 #define SIZE(obj) oe_size(&to_pack, obj)
173 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
174 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
175 #define DELTA(obj) oe_delta(&to_pack, obj)
176 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
177 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
178 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
179 #define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
180 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
181 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
182 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
184 static const char *pack_usage
[] = {
185 N_("git pack-objects --stdout [<options>] [< <ref-list> | < <object-list>]"),
186 N_("git pack-objects [<options>] <base-name> [< <ref-list> | < <object-list>]"),
190 static struct pack_idx_entry
**written_list
;
191 static uint32_t nr_result
, nr_written
, nr_seen
;
192 static struct bitmap_index
*bitmap_git
;
193 static uint32_t write_layer
;
195 static int non_empty
;
196 static int reuse_delta
= 1, reuse_object
= 1;
197 static int keep_unreachable
, unpack_unreachable
, include_tag
;
198 static timestamp_t unpack_unreachable_expiration
;
199 static int pack_loose_unreachable
;
201 static timestamp_t cruft_expiration
;
203 static int have_non_local_packs
;
204 static int incremental
;
205 static int ignore_packed_keep_on_disk
;
206 static int ignore_packed_keep_in_core
;
207 static int allow_ofs_delta
;
208 static struct pack_idx_option pack_idx_opts
;
209 static const char *base_name
;
210 static int progress
= 1;
211 static int window
= 10;
212 static unsigned long pack_size_limit
;
213 static int depth
= 50;
214 static int delta_search_threads
;
215 static int pack_to_stdout
;
218 static int num_preferred_base
;
219 static struct progress
*progress_state
;
221 static struct bitmapped_pack
*reuse_packfiles
;
222 static size_t reuse_packfiles_nr
;
223 static size_t reuse_packfiles_used_nr
;
224 static uint32_t reuse_packfile_objects
;
225 static struct bitmap
*reuse_packfile_bitmap
;
227 static int use_bitmap_index_default
= 1;
228 static int use_bitmap_index
= -1;
233 } allow_pack_reuse
= SINGLE_PACK_REUSE
;
235 WRITE_BITMAP_FALSE
= 0,
238 } write_bitmap_index
;
239 static uint16_t write_bitmap_options
= BITMAP_OPT_HASH_CACHE
;
241 static int exclude_promisor_objects
;
243 static int use_delta_islands
;
245 static unsigned long delta_cache_size
= 0;
246 static unsigned long max_delta_cache_size
= DEFAULT_DELTA_CACHE_SIZE
;
247 static unsigned long cache_max_small_delta_size
= 1000;
249 static unsigned long window_memory_limit
= 0;
251 static struct string_list uri_protocols
= STRING_LIST_INIT_NODUP
;
253 enum missing_action
{
254 MA_ERROR
= 0, /* fail if any missing objects are encountered */
255 MA_ALLOW_ANY
, /* silently allow ALL missing objects */
256 MA_ALLOW_PROMISOR
, /* silently allow all missing PROMISOR objects */
258 static enum missing_action arg_missing_action
;
259 static show_object_fn fn_show_object
;
261 struct configured_exclusion
{
262 struct oidmap_entry e
;
266 static struct oidmap configured_exclusions
;
268 static struct oidset excluded_by_config
;
273 static uint32_t written
, written_delta
;
274 static uint32_t reused
, reused_delta
;
279 static struct commit
**indexed_commits
;
280 static unsigned int indexed_commits_nr
;
281 static unsigned int indexed_commits_alloc
;
283 static void index_commit_for_bitmap(struct commit
*commit
)
285 if (indexed_commits_nr
>= indexed_commits_alloc
) {
286 indexed_commits_alloc
= (indexed_commits_alloc
+ 32) * 2;
287 REALLOC_ARRAY(indexed_commits
, indexed_commits_alloc
);
290 indexed_commits
[indexed_commits_nr
++] = commit
;
293 static void *get_delta(struct object_entry
*entry
)
295 unsigned long size
, base_size
, delta_size
;
296 void *buf
, *base_buf
, *delta_buf
;
297 enum object_type type
;
299 buf
= repo_read_object_file(the_repository
, &entry
->idx
.oid
, &type
,
302 die(_("unable to read %s"), oid_to_hex(&entry
->idx
.oid
));
303 base_buf
= repo_read_object_file(the_repository
,
304 &DELTA(entry
)->idx
.oid
, &type
,
307 die("unable to read %s",
308 oid_to_hex(&DELTA(entry
)->idx
.oid
));
309 delta_buf
= diff_delta(base_buf
, base_size
,
310 buf
, size
, &delta_size
, 0);
312 * We successfully computed this delta once but dropped it for
313 * memory reasons. Something is very wrong if this time we
314 * recompute and create a different delta.
316 if (!delta_buf
|| delta_size
!= DELTA_SIZE(entry
))
317 BUG("delta size changed");
323 static unsigned long do_compress(void **pptr
, unsigned long size
)
327 unsigned long maxsize
;
329 git_deflate_init(&stream
, pack_compression_level
);
330 maxsize
= git_deflate_bound(&stream
, size
);
333 out
= xmalloc(maxsize
);
337 stream
.avail_in
= size
;
338 stream
.next_out
= out
;
339 stream
.avail_out
= maxsize
;
340 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
342 git_deflate_end(&stream
);
345 return stream
.total_out
;
348 static unsigned long write_large_blob_data(struct git_istream
*st
, struct hashfile
*f
,
349 const struct object_id
*oid
)
352 unsigned char ibuf
[1024 * 16];
353 unsigned char obuf
[1024 * 16];
354 unsigned long olen
= 0;
356 git_deflate_init(&stream
, pack_compression_level
);
361 readlen
= read_istream(st
, ibuf
, sizeof(ibuf
));
363 die(_("unable to read %s"), oid_to_hex(oid
));
365 stream
.next_in
= ibuf
;
366 stream
.avail_in
= readlen
;
367 while ((stream
.avail_in
|| readlen
== 0) &&
368 (zret
== Z_OK
|| zret
== Z_BUF_ERROR
)) {
369 stream
.next_out
= obuf
;
370 stream
.avail_out
= sizeof(obuf
);
371 zret
= git_deflate(&stream
, readlen
? 0 : Z_FINISH
);
372 hashwrite(f
, obuf
, stream
.next_out
- obuf
);
373 olen
+= stream
.next_out
- obuf
;
376 die(_("deflate error (%d)"), zret
);
378 if (zret
!= Z_STREAM_END
)
379 die(_("deflate error (%d)"), zret
);
383 git_deflate_end(&stream
);
388 * we are going to reuse the existing object data as is. make
389 * sure it is not corrupt.
391 static int check_pack_inflate(struct packed_git
*p
,
392 struct pack_window
**w_curs
,
395 unsigned long expect
)
398 unsigned char fakebuf
[4096], *in
;
401 memset(&stream
, 0, sizeof(stream
));
402 git_inflate_init(&stream
);
404 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
406 stream
.next_out
= fakebuf
;
407 stream
.avail_out
= sizeof(fakebuf
);
408 st
= git_inflate(&stream
, Z_FINISH
);
409 offset
+= stream
.next_in
- in
;
410 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
411 git_inflate_end(&stream
);
412 return (st
== Z_STREAM_END
&&
413 stream
.total_out
== expect
&&
414 stream
.total_in
== len
) ? 0 : -1;
417 static void copy_pack_data(struct hashfile
*f
,
418 struct packed_git
*p
,
419 struct pack_window
**w_curs
,
427 in
= use_pack(p
, w_curs
, offset
, &avail
);
429 avail
= (unsigned long)len
;
430 hashwrite(f
, in
, avail
);
436 static inline int oe_size_greater_than(struct packing_data
*pack
,
437 const struct object_entry
*lhs
,
441 return lhs
->size_
> rhs
;
442 if (rhs
< pack
->oe_size_limit
) /* rhs < 2^x <= lhs ? */
444 return oe_get_size_slow(pack
, lhs
) > rhs
;
447 /* Return 0 if we will bust the pack-size limit */
448 static unsigned long write_no_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
449 unsigned long limit
, int usable_delta
)
451 unsigned long size
, datalen
;
452 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
453 dheader
[MAX_PACK_OBJECT_HEADER
];
455 enum object_type type
;
457 struct git_istream
*st
= NULL
;
458 const unsigned hashsz
= the_hash_algo
->rawsz
;
461 if (oe_type(entry
) == OBJ_BLOB
&&
462 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
) &&
463 (st
= open_istream(the_repository
, &entry
->idx
.oid
, &type
,
464 &size
, NULL
)) != NULL
)
467 buf
= repo_read_object_file(the_repository
,
468 &entry
->idx
.oid
, &type
,
471 die(_("unable to read %s"),
472 oid_to_hex(&entry
->idx
.oid
));
475 * make sure no cached delta data remains from a
476 * previous attempt before a pack split occurred.
478 FREE_AND_NULL(entry
->delta_data
);
479 entry
->z_delta_size
= 0;
480 } else if (entry
->delta_data
) {
481 size
= DELTA_SIZE(entry
);
482 buf
= entry
->delta_data
;
483 entry
->delta_data
= NULL
;
484 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
485 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
487 buf
= get_delta(entry
);
488 size
= DELTA_SIZE(entry
);
489 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
490 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
493 if (st
) /* large blob case, just assume we don't compress well */
495 else if (entry
->z_delta_size
)
496 datalen
= entry
->z_delta_size
;
498 datalen
= do_compress(&buf
, size
);
501 * The object header is a byte of 'type' followed by zero or
502 * more bytes of length.
504 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
507 if (type
== OBJ_OFS_DELTA
) {
509 * Deltas with relative base contain an additional
510 * encoding of the relative offset for the delta
511 * base from this object's position in the pack.
513 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
514 unsigned pos
= sizeof(dheader
) - 1;
515 dheader
[pos
] = ofs
& 127;
517 dheader
[--pos
] = 128 | (--ofs
& 127);
518 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
524 hashwrite(f
, header
, hdrlen
);
525 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
526 hdrlen
+= sizeof(dheader
) - pos
;
527 } else if (type
== OBJ_REF_DELTA
) {
529 * Deltas with a base reference contain
530 * additional bytes for the base object ID.
532 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
538 hashwrite(f
, header
, hdrlen
);
539 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
542 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
548 hashwrite(f
, header
, hdrlen
);
551 datalen
= write_large_blob_data(st
, f
, &entry
->idx
.oid
);
554 hashwrite(f
, buf
, datalen
);
558 return hdrlen
+ datalen
;
561 /* Return 0 if we will bust the pack-size limit */
562 static off_t
write_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
563 unsigned long limit
, int usable_delta
)
565 struct packed_git
*p
= IN_PACK(entry
);
566 struct pack_window
*w_curs
= NULL
;
569 enum object_type type
= oe_type(entry
);
571 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
572 dheader
[MAX_PACK_OBJECT_HEADER
];
574 const unsigned hashsz
= the_hash_algo
->rawsz
;
575 unsigned long entry_size
= SIZE(entry
);
578 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
579 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
580 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
583 offset
= entry
->in_pack_offset
;
584 if (offset_to_pack_pos(p
, offset
, &pos
) < 0)
585 die(_("write_reuse_object: could not locate %s, expected at "
586 "offset %"PRIuMAX
" in pack %s"),
587 oid_to_hex(&entry
->idx
.oid
), (uintmax_t)offset
,
589 datalen
= pack_pos_to_offset(p
, pos
+ 1) - offset
;
590 if (!pack_to_stdout
&& p
->index_version
> 1 &&
591 check_pack_crc(p
, &w_curs
, offset
, datalen
,
592 pack_pos_to_index(p
, pos
))) {
593 error(_("bad packed object CRC for %s"),
594 oid_to_hex(&entry
->idx
.oid
));
596 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
599 offset
+= entry
->in_pack_header_size
;
600 datalen
-= entry
->in_pack_header_size
;
602 if (!pack_to_stdout
&& p
->index_version
== 1 &&
603 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry_size
)) {
604 error(_("corrupt packed object for %s"),
605 oid_to_hex(&entry
->idx
.oid
));
607 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
610 if (type
== OBJ_OFS_DELTA
) {
611 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
612 unsigned pos
= sizeof(dheader
) - 1;
613 dheader
[pos
] = ofs
& 127;
615 dheader
[--pos
] = 128 | (--ofs
& 127);
616 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
620 hashwrite(f
, header
, hdrlen
);
621 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
622 hdrlen
+= sizeof(dheader
) - pos
;
624 } else if (type
== OBJ_REF_DELTA
) {
625 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
629 hashwrite(f
, header
, hdrlen
);
630 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
634 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
638 hashwrite(f
, header
, hdrlen
);
640 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
643 return hdrlen
+ datalen
;
646 /* Return 0 if we will bust the pack-size limit */
647 static off_t
write_object(struct hashfile
*f
,
648 struct object_entry
*entry
,
653 int usable_delta
, to_reuse
;
658 /* apply size limit if limited packsize and not first object */
659 if (!pack_size_limit
|| !nr_written
)
661 else if (pack_size_limit
<= write_offset
)
663 * the earlier object did not fit the limit; avoid
664 * mistaking this with unlimited (i.e. limit = 0).
668 limit
= pack_size_limit
- write_offset
;
671 usable_delta
= 0; /* no delta */
672 else if (!pack_size_limit
)
673 usable_delta
= 1; /* unlimited packfile */
674 else if (DELTA(entry
)->idx
.offset
== (off_t
)-1)
675 usable_delta
= 0; /* base was written to another pack */
676 else if (DELTA(entry
)->idx
.offset
)
677 usable_delta
= 1; /* base already exists in this pack */
679 usable_delta
= 0; /* base could end up in another pack */
682 to_reuse
= 0; /* explicit */
683 else if (!IN_PACK(entry
))
684 to_reuse
= 0; /* can't reuse what we don't have */
685 else if (oe_type(entry
) == OBJ_REF_DELTA
||
686 oe_type(entry
) == OBJ_OFS_DELTA
)
687 /* check_object() decided it for us ... */
688 to_reuse
= usable_delta
;
689 /* ... but pack split may override that */
690 else if (oe_type(entry
) != entry
->in_pack_type
)
691 to_reuse
= 0; /* pack has delta which is unusable */
692 else if (DELTA(entry
))
693 to_reuse
= 0; /* we want to pack afresh */
695 to_reuse
= 1; /* we have it in-pack undeltified,
696 * and we do not need to deltify it.
700 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
702 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
710 entry
->idx
.crc32
= crc32_end(f
);
714 enum write_one_status
{
715 WRITE_ONE_SKIP
= -1, /* already written */
716 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
717 WRITE_ONE_WRITTEN
= 1, /* normal */
718 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
721 static enum write_one_status
write_one(struct hashfile
*f
,
722 struct object_entry
*e
,
729 * we set offset to 1 (which is an impossible value) to mark
730 * the fact that this object is involved in "write its base
731 * first before writing a deltified object" recursion.
733 recursing
= (e
->idx
.offset
== 1);
735 warning(_("recursive delta detected for object %s"),
736 oid_to_hex(&e
->idx
.oid
));
737 return WRITE_ONE_RECURSIVE
;
738 } else if (e
->idx
.offset
|| e
->preferred_base
) {
739 /* offset is non zero if object is written already. */
740 return WRITE_ONE_SKIP
;
743 /* if we are deltified, write out base object first. */
745 e
->idx
.offset
= 1; /* now recurse */
746 switch (write_one(f
, DELTA(e
), offset
)) {
747 case WRITE_ONE_RECURSIVE
:
748 /* we cannot depend on this one */
753 case WRITE_ONE_BREAK
:
754 e
->idx
.offset
= recursing
;
755 return WRITE_ONE_BREAK
;
759 e
->idx
.offset
= *offset
;
760 size
= write_object(f
, e
, *offset
);
762 e
->idx
.offset
= recursing
;
763 return WRITE_ONE_BREAK
;
765 written_list
[nr_written
++] = &e
->idx
;
767 /* make sure off_t is sufficiently large not to wrap */
768 if (signed_add_overflows(*offset
, size
))
769 die(_("pack too large for current definition of off_t"));
771 return WRITE_ONE_WRITTEN
;
774 static int mark_tagged(const char *path UNUSED
, const struct object_id
*oid
,
775 int flag UNUSED
, void *cb_data UNUSED
)
777 struct object_id peeled
;
778 struct object_entry
*entry
= packlist_find(&to_pack
, oid
);
782 if (!peel_iterated_oid(oid
, &peeled
)) {
783 entry
= packlist_find(&to_pack
, &peeled
);
790 static inline unsigned char oe_layer(struct packing_data
*pack
,
791 struct object_entry
*e
)
795 return pack
->layer
[e
- pack
->objects
];
798 static inline void add_to_write_order(struct object_entry
**wo
,
800 struct object_entry
*e
)
802 if (e
->filled
|| oe_layer(&to_pack
, e
) != write_layer
)
808 static void add_descendants_to_write_order(struct object_entry
**wo
,
810 struct object_entry
*e
)
812 int add_to_order
= 1;
815 struct object_entry
*s
;
816 /* add this node... */
817 add_to_write_order(wo
, endp
, e
);
818 /* all its siblings... */
819 for (s
= DELTA_SIBLING(e
); s
; s
= DELTA_SIBLING(s
)) {
820 add_to_write_order(wo
, endp
, s
);
823 /* drop down a level to add left subtree nodes if possible */
824 if (DELTA_CHILD(e
)) {
829 /* our sibling might have some children, it is next */
830 if (DELTA_SIBLING(e
)) {
831 e
= DELTA_SIBLING(e
);
834 /* go back to our parent node */
836 while (e
&& !DELTA_SIBLING(e
)) {
837 /* we're on the right side of a subtree, keep
838 * going up until we can go right again */
842 /* done- we hit our original root node */
845 /* pass it off to sibling at this level */
846 e
= DELTA_SIBLING(e
);
851 static void add_family_to_write_order(struct object_entry
**wo
,
853 struct object_entry
*e
)
855 struct object_entry
*root
;
857 for (root
= e
; DELTA(root
); root
= DELTA(root
))
859 add_descendants_to_write_order(wo
, endp
, root
);
862 static void compute_layer_order(struct object_entry
**wo
, unsigned int *wo_end
)
864 unsigned int i
, last_untagged
;
865 struct object_entry
*objects
= to_pack
.objects
;
867 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
868 if (objects
[i
].tagged
)
870 add_to_write_order(wo
, wo_end
, &objects
[i
]);
875 * Then fill all the tagged tips.
877 for (; i
< to_pack
.nr_objects
; i
++) {
878 if (objects
[i
].tagged
)
879 add_to_write_order(wo
, wo_end
, &objects
[i
]);
883 * And then all remaining commits and tags.
885 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
886 if (oe_type(&objects
[i
]) != OBJ_COMMIT
&&
887 oe_type(&objects
[i
]) != OBJ_TAG
)
889 add_to_write_order(wo
, wo_end
, &objects
[i
]);
893 * And then all the trees.
895 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
896 if (oe_type(&objects
[i
]) != OBJ_TREE
)
898 add_to_write_order(wo
, wo_end
, &objects
[i
]);
902 * Finally all the rest in really tight order
904 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
905 if (!objects
[i
].filled
&& oe_layer(&to_pack
, &objects
[i
]) == write_layer
)
906 add_family_to_write_order(wo
, wo_end
, &objects
[i
]);
910 static struct object_entry
**compute_write_order(void)
912 uint32_t max_layers
= 1;
913 unsigned int i
, wo_end
;
915 struct object_entry
**wo
;
916 struct object_entry
*objects
= to_pack
.objects
;
918 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
919 objects
[i
].tagged
= 0;
920 objects
[i
].filled
= 0;
921 SET_DELTA_CHILD(&objects
[i
], NULL
);
922 SET_DELTA_SIBLING(&objects
[i
], NULL
);
926 * Fully connect delta_child/delta_sibling network.
927 * Make sure delta_sibling is sorted in the original
930 for (i
= to_pack
.nr_objects
; i
> 0;) {
931 struct object_entry
*e
= &objects
[--i
];
934 /* Mark me as the first child */
935 e
->delta_sibling_idx
= DELTA(e
)->delta_child_idx
;
936 SET_DELTA_CHILD(DELTA(e
), e
);
940 * Mark objects that are at the tip of tags.
942 for_each_tag_ref(mark_tagged
, NULL
);
944 if (use_delta_islands
) {
945 max_layers
= compute_pack_layers(&to_pack
);
949 ALLOC_ARRAY(wo
, to_pack
.nr_objects
);
952 for (; write_layer
< max_layers
; ++write_layer
)
953 compute_layer_order(wo
, &wo_end
);
955 if (wo_end
!= to_pack
.nr_objects
)
956 die(_("ordered %u objects, expected %"PRIu32
),
957 wo_end
, to_pack
.nr_objects
);
964 * A reused set of objects. All objects in a chunk have the same
965 * relative position in the original packfile and the generated
969 static struct reused_chunk
{
970 /* The offset of the first object of this chunk in the original
973 /* The difference for "original" minus the offset of the first object of
974 * this chunk in the generated packfile. */
977 static int reused_chunks_nr
;
978 static int reused_chunks_alloc
;
980 static void record_reused_object(off_t where
, off_t offset
)
982 if (reused_chunks_nr
&& reused_chunks
[reused_chunks_nr
-1].difference
== offset
)
985 ALLOC_GROW(reused_chunks
, reused_chunks_nr
+ 1,
986 reused_chunks_alloc
);
987 reused_chunks
[reused_chunks_nr
].original
= where
;
988 reused_chunks
[reused_chunks_nr
].difference
= offset
;
993 * Binary search to find the chunk that "where" is in. Note
994 * that we're not looking for an exact match, just the first
995 * chunk that contains it (which implicitly ends at the start
998 static off_t
find_reused_offset(off_t where
)
1000 int lo
= 0, hi
= reused_chunks_nr
;
1002 int mi
= lo
+ ((hi
- lo
) / 2);
1003 if (where
== reused_chunks
[mi
].original
)
1004 return reused_chunks
[mi
].difference
;
1005 if (where
< reused_chunks
[mi
].original
)
1012 * The first chunk starts at zero, so we can't have gone below
1016 return reused_chunks
[lo
-1].difference
;
1019 static void write_reused_pack_one(struct packed_git
*reuse_packfile
,
1020 size_t pos
, struct hashfile
*out
,
1022 struct pack_window
**w_curs
)
1024 off_t offset
, next
, cur
;
1025 enum object_type type
;
1028 offset
= pack_pos_to_offset(reuse_packfile
, pos
);
1029 next
= pack_pos_to_offset(reuse_packfile
, pos
+ 1);
1031 record_reused_object(offset
,
1032 offset
- (hashfile_total(out
) - pack_start
));
1035 type
= unpack_object_header(reuse_packfile
, w_curs
, &cur
, &size
);
1038 if (type
== OBJ_OFS_DELTA
) {
1042 unsigned char header
[MAX_PACK_OBJECT_HEADER
];
1045 base_offset
= get_delta_base(reuse_packfile
, w_curs
, &cur
, type
, offset
);
1046 assert(base_offset
!= 0);
1048 /* Convert to REF_DELTA if we must... */
1049 if (!allow_ofs_delta
) {
1051 struct object_id base_oid
;
1053 if (offset_to_pack_pos(reuse_packfile
, base_offset
, &base_pos
) < 0)
1054 die(_("expected object at offset %"PRIuMAX
" "
1056 (uintmax_t)base_offset
,
1057 reuse_packfile
->pack_name
);
1059 nth_packed_object_id(&base_oid
, reuse_packfile
,
1060 pack_pos_to_index(reuse_packfile
, base_pos
));
1062 len
= encode_in_pack_object_header(header
, sizeof(header
),
1063 OBJ_REF_DELTA
, size
);
1064 hashwrite(out
, header
, len
);
1065 hashwrite(out
, base_oid
.hash
, the_hash_algo
->rawsz
);
1066 copy_pack_data(out
, reuse_packfile
, w_curs
, cur
, next
- cur
);
1070 /* Otherwise see if we need to rewrite the offset... */
1071 fixup
= find_reused_offset(offset
) -
1072 find_reused_offset(base_offset
);
1074 unsigned char ofs_header
[10];
1075 unsigned i
, ofs_len
;
1076 off_t ofs
= offset
- base_offset
- fixup
;
1078 len
= encode_in_pack_object_header(header
, sizeof(header
),
1079 OBJ_OFS_DELTA
, size
);
1081 i
= sizeof(ofs_header
) - 1;
1082 ofs_header
[i
] = ofs
& 127;
1084 ofs_header
[--i
] = 128 | (--ofs
& 127);
1086 ofs_len
= sizeof(ofs_header
) - i
;
1088 hashwrite(out
, header
, len
);
1089 hashwrite(out
, ofs_header
+ sizeof(ofs_header
) - ofs_len
, ofs_len
);
1090 copy_pack_data(out
, reuse_packfile
, w_curs
, cur
, next
- cur
);
1094 /* ...otherwise we have no fixup, and can write it verbatim */
1097 copy_pack_data(out
, reuse_packfile
, w_curs
, offset
, next
- offset
);
1100 static size_t write_reused_pack_verbatim(struct bitmapped_pack
*reuse_packfile
,
1101 struct hashfile
*out
,
1103 struct pack_window
**w_curs
)
1105 size_t pos
= reuse_packfile
->bitmap_pos
;
1108 if (pos
% BITS_IN_EWORD
) {
1109 size_t word_pos
= (pos
/ BITS_IN_EWORD
);
1110 size_t offset
= pos
% BITS_IN_EWORD
;
1112 eword_t word
= reuse_packfile_bitmap
->words
[word_pos
];
1114 if (offset
+ reuse_packfile
->bitmap_nr
< BITS_IN_EWORD
)
1115 last
= offset
+ reuse_packfile
->bitmap_nr
;
1117 last
= BITS_IN_EWORD
;
1119 for (; offset
< last
; offset
++) {
1120 if (word
>> offset
== 0)
1122 if (!bitmap_get(reuse_packfile_bitmap
,
1123 word_pos
* BITS_IN_EWORD
+ offset
))
1127 pos
+= BITS_IN_EWORD
- (pos
% BITS_IN_EWORD
);
1131 * Now we're going to copy as many whole eword_t's as possible.
1132 * "end" is the index of the last whole eword_t we copy, but
1133 * there may be additional bits to process. Those are handled
1134 * individually by write_reused_pack().
1136 * Begin by advancing to the first word boundary in range of the
1137 * bit positions occupied by objects in "reuse_packfile". Then
1138 * pick the last word boundary in the same range. If we have at
1139 * least one word's worth of bits to process, continue on.
1141 end
= reuse_packfile
->bitmap_pos
+ reuse_packfile
->bitmap_nr
;
1142 if (end
% BITS_IN_EWORD
)
1143 end
-= end
% BITS_IN_EWORD
;
1145 return reuse_packfile
->bitmap_pos
/ BITS_IN_EWORD
;
1148 reuse_packfile_bitmap
->words
[pos
/ BITS_IN_EWORD
] == (eword_t
)~0)
1149 pos
+= BITS_IN_EWORD
;
1154 if (reuse_packfile
->bitmap_pos
< pos
) {
1155 off_t pack_start_off
= pack_pos_to_offset(reuse_packfile
->p
, 0);
1156 off_t pack_end_off
= pack_pos_to_offset(reuse_packfile
->p
,
1157 pos
- reuse_packfile
->bitmap_pos
);
1159 written
+= pos
- reuse_packfile
->bitmap_pos
;
1161 /* We're recording one chunk, not one object. */
1162 record_reused_object(pack_start_off
,
1163 pack_start_off
- (hashfile_total(out
) - pack_start
));
1165 copy_pack_data(out
, reuse_packfile
->p
, w_curs
,
1166 pack_start_off
, pack_end_off
- pack_start_off
);
1168 display_progress(progress_state
, written
);
1170 if (pos
% BITS_IN_EWORD
)
1171 BUG("attempted to jump past a word boundary to %"PRIuMAX
,
1173 return pos
/ BITS_IN_EWORD
;
1176 static void write_reused_pack(struct bitmapped_pack
*reuse_packfile
,
1179 size_t i
= reuse_packfile
->bitmap_pos
/ BITS_IN_EWORD
;
1181 off_t pack_start
= hashfile_total(f
) - sizeof(struct pack_header
);
1182 struct pack_window
*w_curs
= NULL
;
1184 if (allow_ofs_delta
)
1185 i
= write_reused_pack_verbatim(reuse_packfile
, f
, pack_start
,
1188 for (; i
< reuse_packfile_bitmap
->word_alloc
; ++i
) {
1189 eword_t word
= reuse_packfile_bitmap
->words
[i
];
1190 size_t pos
= (i
* BITS_IN_EWORD
);
1192 for (offset
= 0; offset
< BITS_IN_EWORD
; ++offset
) {
1193 if ((word
>> offset
) == 0)
1196 offset
+= ewah_bit_ctz64(word
>> offset
);
1197 if (pos
+ offset
< reuse_packfile
->bitmap_pos
)
1199 if (pos
+ offset
>= reuse_packfile
->bitmap_pos
+ reuse_packfile
->bitmap_nr
)
1202 * Can use bit positions directly, even for MIDX
1203 * bitmaps. See comment in try_partial_reuse()
1206 write_reused_pack_one(reuse_packfile
->p
,
1207 pos
+ offset
- reuse_packfile
->bitmap_pos
,
1208 f
, pack_start
, &w_curs
);
1209 display_progress(progress_state
, ++written
);
1214 unuse_pack(&w_curs
);
1217 static void write_excluded_by_configs(void)
1219 struct oidset_iter iter
;
1220 const struct object_id
*oid
;
1222 oidset_iter_init(&excluded_by_config
, &iter
);
1223 while ((oid
= oidset_iter_next(&iter
))) {
1224 struct configured_exclusion
*ex
=
1225 oidmap_get(&configured_exclusions
, oid
);
1228 BUG("configured exclusion wasn't configured");
1229 write_in_full(1, ex
->pack_hash_hex
, strlen(ex
->pack_hash_hex
));
1230 write_in_full(1, " ", 1);
1231 write_in_full(1, ex
->uri
, strlen(ex
->uri
));
1232 write_in_full(1, "\n", 1);
1236 static const char no_split_warning
[] = N_(
1237 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
1240 static void write_pack_file(void)
1245 uint32_t nr_remaining
= nr_result
;
1246 time_t last_mtime
= 0;
1247 struct object_entry
**write_order
;
1249 if (progress
> pack_to_stdout
)
1250 progress_state
= start_progress(_("Writing objects"), nr_result
);
1251 ALLOC_ARRAY(written_list
, to_pack
.nr_objects
);
1252 write_order
= compute_write_order();
1255 unsigned char hash
[GIT_MAX_RAWSZ
];
1256 char *pack_tmp_name
= NULL
;
1259 f
= hashfd_throughput(1, "<stdout>", progress_state
);
1261 f
= create_tmp_packfile(&pack_tmp_name
);
1263 offset
= write_pack_header(f
, nr_remaining
);
1265 if (reuse_packfiles_nr
) {
1266 assert(pack_to_stdout
);
1267 for (j
= 0; j
< reuse_packfiles_nr
; j
++) {
1268 reused_chunks_nr
= 0;
1269 write_reused_pack(&reuse_packfiles
[j
], f
);
1270 if (reused_chunks_nr
)
1271 reuse_packfiles_used_nr
++;
1273 offset
= hashfile_total(f
);
1277 for (; i
< to_pack
.nr_objects
; i
++) {
1278 struct object_entry
*e
= write_order
[i
];
1279 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
1281 display_progress(progress_state
, written
);
1284 if (pack_to_stdout
) {
1286 * We never fsync when writing to stdout since we may
1287 * not be writing to an actual pack file. For instance,
1288 * the upload-pack code passes a pipe here. Calling
1289 * fsync on a pipe results in unnecessary
1290 * synchronization with the reader on some platforms.
1292 finalize_hashfile(f
, hash
, FSYNC_COMPONENT_NONE
,
1293 CSUM_HASH_IN_STREAM
| CSUM_CLOSE
);
1294 } else if (nr_written
== nr_remaining
) {
1295 finalize_hashfile(f
, hash
, FSYNC_COMPONENT_PACK
,
1296 CSUM_HASH_IN_STREAM
| CSUM_FSYNC
| CSUM_CLOSE
);
1299 * If we wrote the wrong number of entries in the
1300 * header, rewrite it like in fast-import.
1303 int fd
= finalize_hashfile(f
, hash
, FSYNC_COMPONENT_PACK
, 0);
1304 fixup_pack_header_footer(fd
, hash
, pack_tmp_name
,
1305 nr_written
, hash
, offset
);
1307 if (write_bitmap_index
) {
1308 if (write_bitmap_index
!= WRITE_BITMAP_QUIET
)
1309 warning(_(no_split_warning
));
1310 write_bitmap_index
= 0;
1314 if (!pack_to_stdout
) {
1316 struct strbuf tmpname
= STRBUF_INIT
;
1317 char *idx_tmp_name
= NULL
;
1320 * Packs are runtime accessed in their mtime
1321 * order since newer packs are more likely to contain
1322 * younger objects. So if we are creating multiple
1323 * packs then we should modify the mtime of later ones
1324 * to preserve this property.
1326 if (stat(pack_tmp_name
, &st
) < 0) {
1327 warning_errno(_("failed to stat %s"), pack_tmp_name
);
1328 } else if (!last_mtime
) {
1329 last_mtime
= st
.st_mtime
;
1332 utb
.actime
= st
.st_atime
;
1333 utb
.modtime
= --last_mtime
;
1334 if (utime(pack_tmp_name
, &utb
) < 0)
1335 warning_errno(_("failed utime() on %s"), pack_tmp_name
);
1338 strbuf_addf(&tmpname
, "%s-%s.", base_name
,
1341 if (write_bitmap_index
) {
1342 bitmap_writer_set_checksum(hash
);
1343 bitmap_writer_build_type_index(
1344 &to_pack
, written_list
, nr_written
);
1348 pack_idx_opts
.flags
|= WRITE_MTIMES
;
1350 stage_tmp_packfiles(&tmpname
, pack_tmp_name
,
1351 written_list
, nr_written
,
1352 &to_pack
, &pack_idx_opts
, hash
,
1355 if (write_bitmap_index
) {
1356 size_t tmpname_len
= tmpname
.len
;
1358 strbuf_addstr(&tmpname
, "bitmap");
1359 stop_progress(&progress_state
);
1361 bitmap_writer_show_progress(progress
);
1362 bitmap_writer_select_commits(indexed_commits
, indexed_commits_nr
, -1);
1363 if (bitmap_writer_build(&to_pack
) < 0)
1364 die(_("failed to write bitmap index"));
1365 bitmap_writer_finish(written_list
, nr_written
,
1366 tmpname
.buf
, write_bitmap_options
);
1367 write_bitmap_index
= 0;
1368 strbuf_setlen(&tmpname
, tmpname_len
);
1371 rename_tmp_packfile_idx(&tmpname
, &idx_tmp_name
);
1374 strbuf_release(&tmpname
);
1375 free(pack_tmp_name
);
1376 puts(hash_to_hex(hash
));
1379 /* mark written objects as written to previous pack */
1380 for (j
= 0; j
< nr_written
; j
++) {
1381 written_list
[j
]->offset
= (off_t
)-1;
1383 nr_remaining
-= nr_written
;
1384 } while (nr_remaining
&& i
< to_pack
.nr_objects
);
1388 stop_progress(&progress_state
);
1389 if (written
!= nr_result
)
1390 die(_("wrote %"PRIu32
" objects while expecting %"PRIu32
),
1391 written
, nr_result
);
1392 trace2_data_intmax("pack-objects", the_repository
,
1393 "write_pack_file/wrote", nr_result
);
1396 static int no_try_delta(const char *path
)
1398 static struct attr_check
*check
;
1401 check
= attr_check_initl("delta", NULL
);
1402 git_check_attr(the_repository
->index
, path
, check
);
1403 if (ATTR_FALSE(check
->items
[0].value
))
1409 * When adding an object, check whether we have already added it
1410 * to our packing list. If so, we can skip. However, if we are
1411 * being asked to excludei t, but the previous mention was to include
1412 * it, make sure to adjust its flags and tweak our numbers accordingly.
1414 * As an optimization, we pass out the index position where we would have
1415 * found the item, since that saves us from having to look it up again a
1416 * few lines later when we want to add the new entry.
1418 static int have_duplicate_entry(const struct object_id
*oid
,
1421 struct object_entry
*entry
;
1423 if (reuse_packfile_bitmap
&&
1424 bitmap_walk_contains(bitmap_git
, reuse_packfile_bitmap
, oid
))
1427 entry
= packlist_find(&to_pack
, oid
);
1432 if (!entry
->preferred_base
)
1434 entry
->preferred_base
= 1;
1440 static int want_found_object(const struct object_id
*oid
, int exclude
,
1441 struct packed_git
*p
)
1448 if (!is_pack_valid(p
))
1452 * When asked to do --local (do not include an object that appears in a
1453 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1454 * an object that appears in a pack marked with .keep), finding a pack
1455 * that matches the criteria is sufficient for us to decide to omit it.
1456 * However, even if this pack does not satisfy the criteria, we need to
1457 * make sure no copy of this object appears in _any_ pack that makes us
1458 * to omit the object, so we need to check all the packs.
1460 * We can however first check whether these options can possibly matter;
1461 * if they do not matter we know we want the object in generated pack.
1462 * Otherwise, we signal "-1" at the end to tell the caller that we do
1463 * not know either way, and it needs to check more packs.
1467 * Objects in packs borrowed from elsewhere are discarded regardless of
1468 * if they appear in other packs that weren't borrowed.
1470 if (local
&& !p
->pack_local
)
1474 * Then handle .keep first, as we have a fast(er) path there.
1476 if (ignore_packed_keep_on_disk
|| ignore_packed_keep_in_core
) {
1478 * Set the flags for the kept-pack cache to be the ones we want
1481 * That is, if we are ignoring objects in on-disk keep packs,
1482 * then we want to search through the on-disk keep and ignore
1486 if (ignore_packed_keep_on_disk
)
1487 flags
|= ON_DISK_KEEP_PACKS
;
1488 if (ignore_packed_keep_in_core
)
1489 flags
|= IN_CORE_KEEP_PACKS
;
1491 if (ignore_packed_keep_on_disk
&& p
->pack_keep
)
1493 if (ignore_packed_keep_in_core
&& p
->pack_keep_in_core
)
1495 if (has_object_kept_pack(oid
, flags
))
1500 * At this point we know definitively that either we don't care about
1501 * keep-packs, or the object is not in one. Keep checking other
1504 if (!local
|| !have_non_local_packs
)
1507 /* we don't know yet; keep looking for more packs */
1511 static int want_object_in_pack_one(struct packed_git
*p
,
1512 const struct object_id
*oid
,
1514 struct packed_git
**found_pack
,
1515 off_t
*found_offset
)
1519 if (p
== *found_pack
)
1520 offset
= *found_offset
;
1522 offset
= find_pack_entry_one(oid
->hash
, p
);
1526 if (!is_pack_valid(p
))
1528 *found_offset
= offset
;
1531 return want_found_object(oid
, exclude
, p
);
1537 * Check whether we want the object in the pack (e.g., we do not want
1538 * objects found in non-local stores if the "--local" option was used).
1540 * If the caller already knows an existing pack it wants to take the object
1541 * from, that is passed in *found_pack and *found_offset; otherwise this
1542 * function finds if there is any pack that has the object and returns the pack
1543 * and its offset in these variables.
1545 static int want_object_in_pack(const struct object_id
*oid
,
1547 struct packed_git
**found_pack
,
1548 off_t
*found_offset
)
1551 struct list_head
*pos
;
1552 struct multi_pack_index
*m
;
1554 if (!exclude
&& local
&& has_loose_object_nonlocal(oid
))
1558 * If we already know the pack object lives in, start checks from that
1559 * pack - in the usual case when neither --local was given nor .keep files
1560 * are present we will determine the answer right now.
1563 want
= want_found_object(oid
, exclude
, *found_pack
);
1571 for (m
= get_multi_pack_index(the_repository
); m
; m
= m
->next
) {
1572 struct pack_entry e
;
1573 if (fill_midx_entry(the_repository
, oid
, &e
, m
)) {
1574 want
= want_object_in_pack_one(e
.p
, oid
, exclude
, found_pack
, found_offset
);
1580 list_for_each(pos
, get_packed_git_mru(the_repository
)) {
1581 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1582 want
= want_object_in_pack_one(p
, oid
, exclude
, found_pack
, found_offset
);
1583 if (!exclude
&& want
> 0)
1585 get_packed_git_mru(the_repository
));
1590 if (uri_protocols
.nr
) {
1591 struct configured_exclusion
*ex
=
1592 oidmap_get(&configured_exclusions
, oid
);
1597 for (i
= 0; i
< uri_protocols
.nr
; i
++) {
1598 if (skip_prefix(ex
->uri
,
1599 uri_protocols
.items
[i
].string
,
1602 oidset_insert(&excluded_by_config
, oid
);
1612 static struct object_entry
*create_object_entry(const struct object_id
*oid
,
1613 enum object_type type
,
1617 struct packed_git
*found_pack
,
1620 struct object_entry
*entry
;
1622 entry
= packlist_alloc(&to_pack
, oid
);
1624 oe_set_type(entry
, type
);
1626 entry
->preferred_base
= 1;
1630 oe_set_in_pack(&to_pack
, entry
, found_pack
);
1631 entry
->in_pack_offset
= found_offset
;
1634 entry
->no_try_delta
= no_try_delta
;
1639 static const char no_closure_warning
[] = N_(
1640 "disabling bitmap writing, as some objects are not being packed"
1643 static int add_object_entry(const struct object_id
*oid
, enum object_type type
,
1644 const char *name
, int exclude
)
1646 struct packed_git
*found_pack
= NULL
;
1647 off_t found_offset
= 0;
1649 display_progress(progress_state
, ++nr_seen
);
1651 if (have_duplicate_entry(oid
, exclude
))
1654 if (!want_object_in_pack(oid
, exclude
, &found_pack
, &found_offset
)) {
1655 /* The pack is missing an object, so it will not have closure */
1656 if (write_bitmap_index
) {
1657 if (write_bitmap_index
!= WRITE_BITMAP_QUIET
)
1658 warning(_(no_closure_warning
));
1659 write_bitmap_index
= 0;
1664 create_object_entry(oid
, type
, pack_name_hash(name
),
1665 exclude
, name
&& no_try_delta(name
),
1666 found_pack
, found_offset
);
1670 static int add_object_entry_from_bitmap(const struct object_id
*oid
,
1671 enum object_type type
,
1672 int flags UNUSED
, uint32_t name_hash
,
1673 struct packed_git
*pack
, off_t offset
)
1675 display_progress(progress_state
, ++nr_seen
);
1677 if (have_duplicate_entry(oid
, 0))
1680 if (!want_object_in_pack(oid
, 0, &pack
, &offset
))
1683 create_object_entry(oid
, type
, name_hash
, 0, 0, pack
, offset
);
1687 struct pbase_tree_cache
{
1688 struct object_id oid
;
1692 unsigned long tree_size
;
1695 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1696 static int pbase_tree_cache_ix(const struct object_id
*oid
)
1698 return oid
->hash
[0] % ARRAY_SIZE(pbase_tree_cache
);
1700 static int pbase_tree_cache_ix_incr(int ix
)
1702 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1705 static struct pbase_tree
{
1706 struct pbase_tree
*next
;
1707 /* This is a phony "cache" entry; we are not
1708 * going to evict it or find it through _get()
1709 * mechanism -- this is for the toplevel node that
1710 * would almost always change with any commit.
1712 struct pbase_tree_cache pcache
;
1715 static struct pbase_tree_cache
*pbase_tree_get(const struct object_id
*oid
)
1717 struct pbase_tree_cache
*ent
, *nent
;
1720 enum object_type type
;
1722 int my_ix
= pbase_tree_cache_ix(oid
);
1723 int available_ix
= -1;
1725 /* pbase-tree-cache acts as a limited hashtable.
1726 * your object will be found at your index or within a few
1727 * slots after that slot if it is cached.
1729 for (neigh
= 0; neigh
< 8; neigh
++) {
1730 ent
= pbase_tree_cache
[my_ix
];
1731 if (ent
&& oideq(&ent
->oid
, oid
)) {
1735 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1736 ((0 <= available_ix
) &&
1737 (!ent
&& pbase_tree_cache
[available_ix
])))
1738 available_ix
= my_ix
;
1741 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1744 /* Did not find one. Either we got a bogus request or
1745 * we need to read and perhaps cache.
1747 data
= repo_read_object_file(the_repository
, oid
, &type
, &size
);
1750 if (type
!= OBJ_TREE
) {
1755 /* We need to either cache or return a throwaway copy */
1757 if (available_ix
< 0)
1760 ent
= pbase_tree_cache
[available_ix
];
1761 my_ix
= available_ix
;
1765 nent
= xmalloc(sizeof(*nent
));
1766 nent
->temporary
= (available_ix
< 0);
1769 /* evict and reuse */
1770 free(ent
->tree_data
);
1773 oidcpy(&nent
->oid
, oid
);
1774 nent
->tree_data
= data
;
1775 nent
->tree_size
= size
;
1777 if (!nent
->temporary
)
1778 pbase_tree_cache
[my_ix
] = nent
;
1782 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1784 if (!cache
->temporary
) {
1788 free(cache
->tree_data
);
1792 static size_t name_cmp_len(const char *name
)
1794 return strcspn(name
, "\n/");
1797 static void add_pbase_object(struct tree_desc
*tree
,
1800 const char *fullname
)
1802 struct name_entry entry
;
1805 while (tree_entry(tree
,&entry
)) {
1806 if (S_ISGITLINK(entry
.mode
))
1808 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1809 memcmp(name
, entry
.path
, cmplen
);
1814 if (name
[cmplen
] != '/') {
1815 add_object_entry(&entry
.oid
,
1816 object_type(entry
.mode
),
1820 if (S_ISDIR(entry
.mode
)) {
1821 struct tree_desc sub
;
1822 struct pbase_tree_cache
*tree
;
1823 const char *down
= name
+cmplen
+1;
1824 size_t downlen
= name_cmp_len(down
);
1826 tree
= pbase_tree_get(&entry
.oid
);
1829 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1831 add_pbase_object(&sub
, down
, downlen
, fullname
);
1832 pbase_tree_put(tree
);
1837 static unsigned *done_pbase_paths
;
1838 static int done_pbase_paths_num
;
1839 static int done_pbase_paths_alloc
;
1840 static int done_pbase_path_pos(unsigned hash
)
1843 int hi
= done_pbase_paths_num
;
1845 int mi
= lo
+ (hi
- lo
) / 2;
1846 if (done_pbase_paths
[mi
] == hash
)
1848 if (done_pbase_paths
[mi
] < hash
)
1856 static int check_pbase_path(unsigned hash
)
1858 int pos
= done_pbase_path_pos(hash
);
1862 ALLOC_GROW(done_pbase_paths
,
1863 done_pbase_paths_num
+ 1,
1864 done_pbase_paths_alloc
);
1865 done_pbase_paths_num
++;
1866 if (pos
< done_pbase_paths_num
)
1867 MOVE_ARRAY(done_pbase_paths
+ pos
+ 1, done_pbase_paths
+ pos
,
1868 done_pbase_paths_num
- pos
- 1);
1869 done_pbase_paths
[pos
] = hash
;
1873 static void add_preferred_base_object(const char *name
)
1875 struct pbase_tree
*it
;
1877 unsigned hash
= pack_name_hash(name
);
1879 if (!num_preferred_base
|| check_pbase_path(hash
))
1882 cmplen
= name_cmp_len(name
);
1883 for (it
= pbase_tree
; it
; it
= it
->next
) {
1885 add_object_entry(&it
->pcache
.oid
, OBJ_TREE
, NULL
, 1);
1888 struct tree_desc tree
;
1889 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1890 add_pbase_object(&tree
, name
, cmplen
, name
);
1895 static void add_preferred_base(struct object_id
*oid
)
1897 struct pbase_tree
*it
;
1900 struct object_id tree_oid
;
1902 if (window
<= num_preferred_base
++)
1905 data
= read_object_with_reference(the_repository
, oid
,
1906 OBJ_TREE
, &size
, &tree_oid
);
1910 for (it
= pbase_tree
; it
; it
= it
->next
) {
1911 if (oideq(&it
->pcache
.oid
, &tree_oid
)) {
1917 CALLOC_ARRAY(it
, 1);
1918 it
->next
= pbase_tree
;
1921 oidcpy(&it
->pcache
.oid
, &tree_oid
);
1922 it
->pcache
.tree_data
= data
;
1923 it
->pcache
.tree_size
= size
;
1926 static void cleanup_preferred_base(void)
1928 struct pbase_tree
*it
;
1934 struct pbase_tree
*tmp
= it
;
1936 free(tmp
->pcache
.tree_data
);
1940 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1941 if (!pbase_tree_cache
[i
])
1943 free(pbase_tree_cache
[i
]->tree_data
);
1944 FREE_AND_NULL(pbase_tree_cache
[i
]);
1947 FREE_AND_NULL(done_pbase_paths
);
1948 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1952 * Return 1 iff the object specified by "delta" can be sent
1953 * literally as a delta against the base in "base_sha1". If
1954 * so, then *base_out will point to the entry in our packing
1955 * list, or NULL if we must use the external-base list.
1957 * Depth value does not matter - find_deltas() will
1958 * never consider reused delta as the base object to
1959 * deltify other objects against, in order to avoid
1962 static int can_reuse_delta(const struct object_id
*base_oid
,
1963 struct object_entry
*delta
,
1964 struct object_entry
**base_out
)
1966 struct object_entry
*base
;
1969 * First see if we're already sending the base (or it's explicitly in
1970 * our "excluded" list).
1972 base
= packlist_find(&to_pack
, base_oid
);
1974 if (!in_same_island(&delta
->idx
.oid
, &base
->idx
.oid
))
1981 * Otherwise, reachability bitmaps may tell us if the receiver has it,
1982 * even if it was buried too deep in history to make it into the
1985 if (thin
&& bitmap_has_oid_in_uninteresting(bitmap_git
, base_oid
)) {
1986 if (use_delta_islands
) {
1987 if (!in_same_island(&delta
->idx
.oid
, base_oid
))
1997 static void prefetch_to_pack(uint32_t object_index_start
) {
1998 struct oid_array to_fetch
= OID_ARRAY_INIT
;
2001 for (i
= object_index_start
; i
< to_pack
.nr_objects
; i
++) {
2002 struct object_entry
*entry
= to_pack
.objects
+ i
;
2004 if (!oid_object_info_extended(the_repository
,
2007 OBJECT_INFO_FOR_PREFETCH
))
2009 oid_array_append(&to_fetch
, &entry
->idx
.oid
);
2011 promisor_remote_get_direct(the_repository
,
2012 to_fetch
.oid
, to_fetch
.nr
);
2013 oid_array_clear(&to_fetch
);
2016 static void check_object(struct object_entry
*entry
, uint32_t object_index
)
2018 unsigned long canonical_size
;
2019 enum object_type type
;
2020 struct object_info oi
= {.typep
= &type
, .sizep
= &canonical_size
};
2022 if (IN_PACK(entry
)) {
2023 struct packed_git
*p
= IN_PACK(entry
);
2024 struct pack_window
*w_curs
= NULL
;
2026 struct object_id base_ref
;
2027 struct object_entry
*base_entry
;
2028 unsigned long used
, used_0
;
2029 unsigned long avail
;
2031 unsigned char *buf
, c
;
2032 enum object_type type
;
2033 unsigned long in_pack_size
;
2035 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
2038 * We want in_pack_type even if we do not reuse delta
2039 * since non-delta representations could still be reused.
2041 used
= unpack_object_header_buffer(buf
, avail
,
2048 BUG("invalid type %d", type
);
2049 entry
->in_pack_type
= type
;
2052 * Determine if this is a delta and if so whether we can
2053 * reuse it or not. Otherwise let's find out as cheaply as
2054 * possible what the actual type and size for this object is.
2056 switch (entry
->in_pack_type
) {
2058 /* Not a delta hence we've already got all we need. */
2059 oe_set_type(entry
, entry
->in_pack_type
);
2060 SET_SIZE(entry
, in_pack_size
);
2061 entry
->in_pack_header_size
= used
;
2062 if (oe_type(entry
) < OBJ_COMMIT
|| oe_type(entry
) > OBJ_BLOB
)
2064 unuse_pack(&w_curs
);
2067 if (reuse_delta
&& !entry
->preferred_base
) {
2069 use_pack(p
, &w_curs
,
2070 entry
->in_pack_offset
+ used
,
2074 entry
->in_pack_header_size
= used
+ the_hash_algo
->rawsz
;
2077 buf
= use_pack(p
, &w_curs
,
2078 entry
->in_pack_offset
+ used
, NULL
);
2084 if (!ofs
|| MSB(ofs
, 7)) {
2085 error(_("delta base offset overflow in pack for %s"),
2086 oid_to_hex(&entry
->idx
.oid
));
2090 ofs
= (ofs
<< 7) + (c
& 127);
2092 ofs
= entry
->in_pack_offset
- ofs
;
2093 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
2094 error(_("delta base offset out of bound for %s"),
2095 oid_to_hex(&entry
->idx
.oid
));
2098 if (reuse_delta
&& !entry
->preferred_base
) {
2100 if (offset_to_pack_pos(p
, ofs
, &pos
) < 0)
2102 if (!nth_packed_object_id(&base_ref
, p
,
2103 pack_pos_to_index(p
, pos
)))
2106 entry
->in_pack_header_size
= used
+ used_0
;
2111 can_reuse_delta(&base_ref
, entry
, &base_entry
)) {
2112 oe_set_type(entry
, entry
->in_pack_type
);
2113 SET_SIZE(entry
, in_pack_size
); /* delta size */
2114 SET_DELTA_SIZE(entry
, in_pack_size
);
2117 SET_DELTA(entry
, base_entry
);
2118 entry
->delta_sibling_idx
= base_entry
->delta_child_idx
;
2119 SET_DELTA_CHILD(base_entry
, entry
);
2121 SET_DELTA_EXT(entry
, &base_ref
);
2124 unuse_pack(&w_curs
);
2128 if (oe_type(entry
)) {
2132 * This must be a delta and we already know what the
2133 * final object type is. Let's extract the actual
2134 * object size from the delta header.
2136 delta_pos
= entry
->in_pack_offset
+ entry
->in_pack_header_size
;
2137 canonical_size
= get_size_from_delta(p
, &w_curs
, delta_pos
);
2138 if (canonical_size
== 0)
2140 SET_SIZE(entry
, canonical_size
);
2141 unuse_pack(&w_curs
);
2146 * No choice but to fall back to the recursive delta walk
2147 * with oid_object_info() to find about the object type
2151 unuse_pack(&w_curs
);
2154 if (oid_object_info_extended(the_repository
, &entry
->idx
.oid
, &oi
,
2155 OBJECT_INFO_SKIP_FETCH_OBJECT
| OBJECT_INFO_LOOKUP_REPLACE
) < 0) {
2156 if (repo_has_promisor_remote(the_repository
)) {
2157 prefetch_to_pack(object_index
);
2158 if (oid_object_info_extended(the_repository
, &entry
->idx
.oid
, &oi
,
2159 OBJECT_INFO_SKIP_FETCH_OBJECT
| OBJECT_INFO_LOOKUP_REPLACE
) < 0)
2165 oe_set_type(entry
, type
);
2166 if (entry
->type_valid
) {
2167 SET_SIZE(entry
, canonical_size
);
2170 * Bad object type is checked in prepare_pack(). This is
2171 * to permit a missing preferred base object to be ignored
2172 * as a preferred base. Doing so can result in a larger
2173 * pack file, but the transfer will still take place.
2178 static int pack_offset_sort(const void *_a
, const void *_b
)
2180 const struct object_entry
*a
= *(struct object_entry
**)_a
;
2181 const struct object_entry
*b
= *(struct object_entry
**)_b
;
2182 const struct packed_git
*a_in_pack
= IN_PACK(a
);
2183 const struct packed_git
*b_in_pack
= IN_PACK(b
);
2185 /* avoid filesystem trashing with loose objects */
2186 if (!a_in_pack
&& !b_in_pack
)
2187 return oidcmp(&a
->idx
.oid
, &b
->idx
.oid
);
2189 if (a_in_pack
< b_in_pack
)
2191 if (a_in_pack
> b_in_pack
)
2193 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
2194 (a
->in_pack_offset
> b
->in_pack_offset
);
2198 * Drop an on-disk delta we were planning to reuse. Naively, this would
2199 * just involve blanking out the "delta" field, but we have to deal
2200 * with some extra book-keeping:
2202 * 1. Removing ourselves from the delta_sibling linked list.
2204 * 2. Updating our size/type to the non-delta representation. These were
2205 * either not recorded initially (size) or overwritten with the delta type
2206 * (type) when check_object() decided to reuse the delta.
2208 * 3. Resetting our delta depth, as we are now a base object.
2210 static void drop_reused_delta(struct object_entry
*entry
)
2212 unsigned *idx
= &to_pack
.objects
[entry
->delta_idx
- 1].delta_child_idx
;
2213 struct object_info oi
= OBJECT_INFO_INIT
;
2214 enum object_type type
;
2218 struct object_entry
*oe
= &to_pack
.objects
[*idx
- 1];
2221 *idx
= oe
->delta_sibling_idx
;
2223 idx
= &oe
->delta_sibling_idx
;
2225 SET_DELTA(entry
, NULL
);
2230 if (packed_object_info(the_repository
, IN_PACK(entry
), entry
->in_pack_offset
, &oi
) < 0) {
2232 * We failed to get the info from this pack for some reason;
2233 * fall back to oid_object_info, which may find another copy.
2234 * And if that fails, the error will be recorded in oe_type(entry)
2235 * and dealt with in prepare_pack().
2238 oid_object_info(the_repository
, &entry
->idx
.oid
, &size
));
2240 oe_set_type(entry
, type
);
2242 SET_SIZE(entry
, size
);
2246 * Follow the chain of deltas from this entry onward, throwing away any links
2247 * that cause us to hit a cycle (as determined by the DFS state flags in
2250 * We also detect too-long reused chains that would violate our --depth
2253 static void break_delta_chains(struct object_entry
*entry
)
2256 * The actual depth of each object we will write is stored as an int,
2257 * as it cannot exceed our int "depth" limit. But before we break
2258 * changes based no that limit, we may potentially go as deep as the
2259 * number of objects, which is elsewhere bounded to a uint32_t.
2261 uint32_t total_depth
;
2262 struct object_entry
*cur
, *next
;
2264 for (cur
= entry
, total_depth
= 0;
2266 cur
= DELTA(cur
), total_depth
++) {
2267 if (cur
->dfs_state
== DFS_DONE
) {
2269 * We've already seen this object and know it isn't
2270 * part of a cycle. We do need to append its depth
2273 total_depth
+= cur
->depth
;
2278 * We break cycles before looping, so an ACTIVE state (or any
2279 * other cruft which made its way into the state variable)
2282 if (cur
->dfs_state
!= DFS_NONE
)
2283 BUG("confusing delta dfs state in first pass: %d",
2287 * Now we know this is the first time we've seen the object. If
2288 * it's not a delta, we're done traversing, but we'll mark it
2289 * done to save time on future traversals.
2292 cur
->dfs_state
= DFS_DONE
;
2297 * Mark ourselves as active and see if the next step causes
2298 * us to cycle to another active object. It's important to do
2299 * this _before_ we loop, because it impacts where we make the
2300 * cut, and thus how our total_depth counter works.
2301 * E.g., We may see a partial loop like:
2303 * A -> B -> C -> D -> B
2305 * Cutting B->C breaks the cycle. But now the depth of A is
2306 * only 1, and our total_depth counter is at 3. The size of the
2307 * error is always one less than the size of the cycle we
2308 * broke. Commits C and D were "lost" from A's chain.
2310 * If we instead cut D->B, then the depth of A is correct at 3.
2311 * We keep all commits in the chain that we examined.
2313 cur
->dfs_state
= DFS_ACTIVE
;
2314 if (DELTA(cur
)->dfs_state
== DFS_ACTIVE
) {
2315 drop_reused_delta(cur
);
2316 cur
->dfs_state
= DFS_DONE
;
2322 * And now that we've gone all the way to the bottom of the chain, we
2323 * need to clear the active flags and set the depth fields as
2324 * appropriate. Unlike the loop above, which can quit when it drops a
2325 * delta, we need to keep going to look for more depth cuts. So we need
2326 * an extra "next" pointer to keep going after we reset cur->delta.
2328 for (cur
= entry
; cur
; cur
= next
) {
2332 * We should have a chain of zero or more ACTIVE states down to
2333 * a final DONE. We can quit after the DONE, because either it
2334 * has no bases, or we've already handled them in a previous
2337 if (cur
->dfs_state
== DFS_DONE
)
2339 else if (cur
->dfs_state
!= DFS_ACTIVE
)
2340 BUG("confusing delta dfs state in second pass: %d",
2344 * If the total_depth is more than depth, then we need to snip
2345 * the chain into two or more smaller chains that don't exceed
2346 * the maximum depth. Most of the resulting chains will contain
2347 * (depth + 1) entries (i.e., depth deltas plus one base), and
2348 * the last chain (i.e., the one containing entry) will contain
2349 * whatever entries are left over, namely
2350 * (total_depth % (depth + 1)) of them.
2352 * Since we are iterating towards decreasing depth, we need to
2353 * decrement total_depth as we go, and we need to write to the
2354 * entry what its final depth will be after all of the
2355 * snipping. Since we're snipping into chains of length (depth
2356 * + 1) entries, the final depth of an entry will be its
2357 * original depth modulo (depth + 1). Any time we encounter an
2358 * entry whose final depth is supposed to be zero, we snip it
2359 * from its delta base, thereby making it so.
2361 cur
->depth
= (total_depth
--) % (depth
+ 1);
2363 drop_reused_delta(cur
);
2365 cur
->dfs_state
= DFS_DONE
;
2369 static void get_object_details(void)
2372 struct object_entry
**sorted_by_offset
;
2375 progress_state
= start_progress(_("Counting objects"),
2376 to_pack
.nr_objects
);
2378 CALLOC_ARRAY(sorted_by_offset
, to_pack
.nr_objects
);
2379 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
2380 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
2381 QSORT(sorted_by_offset
, to_pack
.nr_objects
, pack_offset_sort
);
2383 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2384 struct object_entry
*entry
= sorted_by_offset
[i
];
2385 check_object(entry
, i
);
2386 if (entry
->type_valid
&&
2387 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
))
2388 entry
->no_try_delta
= 1;
2389 display_progress(progress_state
, i
+ 1);
2391 stop_progress(&progress_state
);
2394 * This must happen in a second pass, since we rely on the delta
2395 * information for the whole list being completed.
2397 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
2398 break_delta_chains(&to_pack
.objects
[i
]);
2400 free(sorted_by_offset
);
2404 * We search for deltas in a list sorted by type, by filename hash, and then
2405 * by size, so that we see progressively smaller and smaller files.
2406 * That's because we prefer deltas to be from the bigger file
2407 * to the smaller -- deletes are potentially cheaper, but perhaps
2408 * more importantly, the bigger file is likely the more recent
2409 * one. The deepest deltas are therefore the oldest objects which are
2410 * less susceptible to be accessed often.
2412 static int type_size_sort(const void *_a
, const void *_b
)
2414 const struct object_entry
*a
= *(struct object_entry
**)_a
;
2415 const struct object_entry
*b
= *(struct object_entry
**)_b
;
2416 const enum object_type a_type
= oe_type(a
);
2417 const enum object_type b_type
= oe_type(b
);
2418 const unsigned long a_size
= SIZE(a
);
2419 const unsigned long b_size
= SIZE(b
);
2421 if (a_type
> b_type
)
2423 if (a_type
< b_type
)
2425 if (a
->hash
> b
->hash
)
2427 if (a
->hash
< b
->hash
)
2429 if (a
->preferred_base
> b
->preferred_base
)
2431 if (a
->preferred_base
< b
->preferred_base
)
2433 if (use_delta_islands
) {
2434 const int island_cmp
= island_delta_cmp(&a
->idx
.oid
, &b
->idx
.oid
);
2438 if (a_size
> b_size
)
2440 if (a_size
< b_size
)
2442 return a
< b
? -1 : (a
> b
); /* newest first */
2446 struct object_entry
*entry
;
2448 struct delta_index
*index
;
2452 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
2453 unsigned long delta_size
)
2455 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
2458 if (delta_size
< cache_max_small_delta_size
)
2461 /* cache delta, if objects are large enough compared to delta size */
2462 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
2468 /* Protect delta_cache_size */
2469 static pthread_mutex_t cache_mutex
;
2470 #define cache_lock() pthread_mutex_lock(&cache_mutex)
2471 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
2474 * Protect object list partitioning (e.g. struct thread_param) and
2477 static pthread_mutex_t progress_mutex
;
2478 #define progress_lock() pthread_mutex_lock(&progress_mutex)
2479 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
2482 * Access to struct object_entry is unprotected since each thread owns
2483 * a portion of the main object list. Just don't access object entries
2484 * ahead in the list because they can be stolen and would need
2485 * progress_mutex for protection.
2488 static inline int oe_size_less_than(struct packing_data
*pack
,
2489 const struct object_entry
*lhs
,
2492 if (lhs
->size_valid
)
2493 return lhs
->size_
< rhs
;
2494 if (rhs
< pack
->oe_size_limit
) /* rhs < 2^x <= lhs ? */
2496 return oe_get_size_slow(pack
, lhs
) < rhs
;
2499 static inline void oe_set_tree_depth(struct packing_data
*pack
,
2500 struct object_entry
*e
,
2501 unsigned int tree_depth
)
2503 if (!pack
->tree_depth
)
2504 CALLOC_ARRAY(pack
->tree_depth
, pack
->nr_alloc
);
2505 pack
->tree_depth
[e
- pack
->objects
] = tree_depth
;
2509 * Return the size of the object without doing any delta
2510 * reconstruction (so non-deltas are true object sizes, but deltas
2511 * return the size of the delta data).
2513 unsigned long oe_get_size_slow(struct packing_data
*pack
,
2514 const struct object_entry
*e
)
2516 struct packed_git
*p
;
2517 struct pack_window
*w_curs
;
2519 enum object_type type
;
2520 unsigned long used
, avail
, size
;
2522 if (e
->type_
!= OBJ_OFS_DELTA
&& e
->type_
!= OBJ_REF_DELTA
) {
2523 packing_data_lock(&to_pack
);
2524 if (oid_object_info(the_repository
, &e
->idx
.oid
, &size
) < 0)
2525 die(_("unable to get size of %s"),
2526 oid_to_hex(&e
->idx
.oid
));
2527 packing_data_unlock(&to_pack
);
2531 p
= oe_in_pack(pack
, e
);
2533 BUG("when e->type is a delta, it must belong to a pack");
2535 packing_data_lock(&to_pack
);
2537 buf
= use_pack(p
, &w_curs
, e
->in_pack_offset
, &avail
);
2538 used
= unpack_object_header_buffer(buf
, avail
, &type
, &size
);
2540 die(_("unable to parse object header of %s"),
2541 oid_to_hex(&e
->idx
.oid
));
2543 unuse_pack(&w_curs
);
2544 packing_data_unlock(&to_pack
);
2548 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
2549 unsigned max_depth
, unsigned long *mem_usage
)
2551 struct object_entry
*trg_entry
= trg
->entry
;
2552 struct object_entry
*src_entry
= src
->entry
;
2553 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
2555 enum object_type type
;
2558 /* Don't bother doing diffs between different types */
2559 if (oe_type(trg_entry
) != oe_type(src_entry
))
2563 * We do not bother to try a delta that we discarded on an
2564 * earlier try, but only when reusing delta data. Note that
2565 * src_entry that is marked as the preferred_base should always
2566 * be considered, as even if we produce a suboptimal delta against
2567 * it, we will still save the transfer cost, as we already know
2568 * the other side has it and we won't send src_entry at all.
2570 if (reuse_delta
&& IN_PACK(trg_entry
) &&
2571 IN_PACK(trg_entry
) == IN_PACK(src_entry
) &&
2572 !src_entry
->preferred_base
&&
2573 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
2574 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
2577 /* Let's not bust the allowed depth. */
2578 if (src
->depth
>= max_depth
)
2581 /* Now some size filtering heuristics. */
2582 trg_size
= SIZE(trg_entry
);
2583 if (!DELTA(trg_entry
)) {
2584 max_size
= trg_size
/2 - the_hash_algo
->rawsz
;
2587 max_size
= DELTA_SIZE(trg_entry
);
2588 ref_depth
= trg
->depth
;
2590 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
2591 (max_depth
- ref_depth
+ 1);
2594 src_size
= SIZE(src_entry
);
2595 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
2596 if (sizediff
>= max_size
)
2598 if (trg_size
< src_size
/ 32)
2601 if (!in_same_island(&trg
->entry
->idx
.oid
, &src
->entry
->idx
.oid
))
2604 /* Load data if not already done */
2606 packing_data_lock(&to_pack
);
2607 trg
->data
= repo_read_object_file(the_repository
,
2608 &trg_entry
->idx
.oid
, &type
,
2610 packing_data_unlock(&to_pack
);
2612 die(_("object %s cannot be read"),
2613 oid_to_hex(&trg_entry
->idx
.oid
));
2615 die(_("object %s inconsistent object length (%"PRIuMAX
" vs %"PRIuMAX
")"),
2616 oid_to_hex(&trg_entry
->idx
.oid
), (uintmax_t)sz
,
2617 (uintmax_t)trg_size
);
2621 packing_data_lock(&to_pack
);
2622 src
->data
= repo_read_object_file(the_repository
,
2623 &src_entry
->idx
.oid
, &type
,
2625 packing_data_unlock(&to_pack
);
2627 if (src_entry
->preferred_base
) {
2628 static int warned
= 0;
2630 warning(_("object %s cannot be read"),
2631 oid_to_hex(&src_entry
->idx
.oid
));
2633 * Those objects are not included in the
2634 * resulting pack. Be resilient and ignore
2635 * them if they can't be read, in case the
2636 * pack could be created nevertheless.
2640 die(_("object %s cannot be read"),
2641 oid_to_hex(&src_entry
->idx
.oid
));
2644 die(_("object %s inconsistent object length (%"PRIuMAX
" vs %"PRIuMAX
")"),
2645 oid_to_hex(&src_entry
->idx
.oid
), (uintmax_t)sz
,
2646 (uintmax_t)src_size
);
2650 src
->index
= create_delta_index(src
->data
, src_size
);
2652 static int warned
= 0;
2654 warning(_("suboptimal pack - out of memory"));
2657 *mem_usage
+= sizeof_delta_index(src
->index
);
2660 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
2664 if (DELTA(trg_entry
)) {
2665 /* Prefer only shallower same-sized deltas. */
2666 if (delta_size
== DELTA_SIZE(trg_entry
) &&
2667 src
->depth
+ 1 >= trg
->depth
) {
2674 * Handle memory allocation outside of the cache
2675 * accounting lock. Compiler will optimize the strangeness
2676 * away when NO_PTHREADS is defined.
2678 free(trg_entry
->delta_data
);
2680 if (trg_entry
->delta_data
) {
2681 delta_cache_size
-= DELTA_SIZE(trg_entry
);
2682 trg_entry
->delta_data
= NULL
;
2684 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
2685 delta_cache_size
+= delta_size
;
2687 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
2693 SET_DELTA(trg_entry
, src_entry
);
2694 SET_DELTA_SIZE(trg_entry
, delta_size
);
2695 trg
->depth
= src
->depth
+ 1;
2700 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
2702 struct object_entry
*child
= DELTA_CHILD(me
);
2705 const unsigned int c
= check_delta_limit(child
, n
+ 1);
2708 child
= DELTA_SIBLING(child
);
2713 static unsigned long free_unpacked(struct unpacked
*n
)
2715 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
2716 free_delta_index(n
->index
);
2719 freed_mem
+= SIZE(n
->entry
);
2720 FREE_AND_NULL(n
->data
);
2727 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
2728 int window
, int depth
, unsigned *processed
)
2730 uint32_t i
, idx
= 0, count
= 0;
2731 struct unpacked
*array
;
2732 unsigned long mem_usage
= 0;
2734 CALLOC_ARRAY(array
, window
);
2737 struct object_entry
*entry
;
2738 struct unpacked
*n
= array
+ idx
;
2739 int j
, max_depth
, best_base
= -1;
2748 if (!entry
->preferred_base
) {
2750 display_progress(progress_state
, *processed
);
2754 mem_usage
-= free_unpacked(n
);
2757 while (window_memory_limit
&&
2758 mem_usage
> window_memory_limit
&&
2760 const uint32_t tail
= (idx
+ window
- count
) % window
;
2761 mem_usage
-= free_unpacked(array
+ tail
);
2765 /* We do not compute delta to *create* objects we are not
2768 if (entry
->preferred_base
)
2772 * If the current object is at pack edge, take the depth the
2773 * objects that depend on the current object into account
2774 * otherwise they would become too deep.
2777 if (DELTA_CHILD(entry
)) {
2778 max_depth
-= check_delta_limit(entry
, 0);
2786 uint32_t other_idx
= idx
+ j
;
2788 if (other_idx
>= window
)
2789 other_idx
-= window
;
2790 m
= array
+ other_idx
;
2793 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
2797 best_base
= other_idx
;
2801 * If we decided to cache the delta data, then it is best
2802 * to compress it right away. First because we have to do
2803 * it anyway, and doing it here while we're threaded will
2804 * save a lot of time in the non threaded write phase,
2805 * as well as allow for caching more deltas within
2806 * the same cache size limit.
2808 * But only if not writing to stdout, since in that case
2809 * the network is most likely throttling writes anyway,
2810 * and therefore it is best to go to the write phase ASAP
2811 * instead, as we can afford spending more time compressing
2812 * between writes at that moment.
2814 if (entry
->delta_data
&& !pack_to_stdout
) {
2817 size
= do_compress(&entry
->delta_data
, DELTA_SIZE(entry
));
2818 if (size
< (1U << OE_Z_DELTA_BITS
)) {
2819 entry
->z_delta_size
= size
;
2821 delta_cache_size
-= DELTA_SIZE(entry
);
2822 delta_cache_size
+= entry
->z_delta_size
;
2825 FREE_AND_NULL(entry
->delta_data
);
2826 entry
->z_delta_size
= 0;
2830 /* if we made n a delta, and if n is already at max
2831 * depth, leaving it in the window is pointless. we
2832 * should evict it first.
2834 if (DELTA(entry
) && max_depth
<= n
->depth
)
2838 * Move the best delta base up in the window, after the
2839 * currently deltified object, to keep it longer. It will
2840 * be the first base object to be attempted next.
2843 struct unpacked swap
= array
[best_base
];
2844 int dist
= (window
+ idx
- best_base
) % window
;
2845 int dst
= best_base
;
2847 int src
= (dst
+ 1) % window
;
2848 array
[dst
] = array
[src
];
2856 if (count
+ 1 < window
)
2862 for (i
= 0; i
< window
; ++i
) {
2863 free_delta_index(array
[i
].index
);
2864 free(array
[i
].data
);
2870 * The main object list is split into smaller lists, each is handed to
2873 * The main thread waits on the condition that (at least) one of the workers
2874 * has stopped working (which is indicated in the .working member of
2875 * struct thread_params).
2877 * When a work thread has completed its work, it sets .working to 0 and
2878 * signals the main thread and waits on the condition that .data_ready
2881 * The main thread steals half of the work from the worker that has
2882 * most work left to hand it to the idle worker.
2885 struct thread_params
{
2887 struct object_entry
**list
;
2894 pthread_mutex_t mutex
;
2895 pthread_cond_t cond
;
2896 unsigned *processed
;
2899 static pthread_cond_t progress_cond
;
2902 * Mutex and conditional variable can't be statically-initialized on Windows.
2904 static void init_threaded_search(void)
2906 pthread_mutex_init(&cache_mutex
, NULL
);
2907 pthread_mutex_init(&progress_mutex
, NULL
);
2908 pthread_cond_init(&progress_cond
, NULL
);
2911 static void cleanup_threaded_search(void)
2913 pthread_cond_destroy(&progress_cond
);
2914 pthread_mutex_destroy(&cache_mutex
);
2915 pthread_mutex_destroy(&progress_mutex
);
2918 static void *threaded_find_deltas(void *arg
)
2920 struct thread_params
*me
= arg
;
2923 while (me
->remaining
) {
2926 find_deltas(me
->list
, &me
->remaining
,
2927 me
->window
, me
->depth
, me
->processed
);
2931 pthread_cond_signal(&progress_cond
);
2935 * We must not set ->data_ready before we wait on the
2936 * condition because the main thread may have set it to 1
2937 * before we get here. In order to be sure that new
2938 * work is available if we see 1 in ->data_ready, it
2939 * was initialized to 0 before this thread was spawned
2940 * and we reset it to 0 right away.
2942 pthread_mutex_lock(&me
->mutex
);
2943 while (!me
->data_ready
)
2944 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2946 pthread_mutex_unlock(&me
->mutex
);
2951 /* leave ->working 1 so that this doesn't get more work assigned */
2955 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2956 int window
, int depth
, unsigned *processed
)
2958 struct thread_params
*p
;
2959 int i
, ret
, active_threads
= 0;
2961 init_threaded_search();
2963 if (delta_search_threads
<= 1) {
2964 find_deltas(list
, &list_size
, window
, depth
, processed
);
2965 cleanup_threaded_search();
2968 if (progress
> pack_to_stdout
)
2969 fprintf_ln(stderr
, _("Delta compression using up to %d threads"),
2970 delta_search_threads
);
2971 CALLOC_ARRAY(p
, delta_search_threads
);
2973 /* Partition the work amongst work threads. */
2974 for (i
= 0; i
< delta_search_threads
; i
++) {
2975 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2977 /* don't use too small segments or no deltas will be found */
2978 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2981 p
[i
].window
= window
;
2983 p
[i
].processed
= processed
;
2985 p
[i
].data_ready
= 0;
2987 /* try to split chunks on "path" boundaries */
2988 while (sub_size
&& sub_size
< list_size
&&
2989 list
[sub_size
]->hash
&&
2990 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2994 p
[i
].list_size
= sub_size
;
2995 p
[i
].remaining
= sub_size
;
2998 list_size
-= sub_size
;
3001 /* Start work threads. */
3002 for (i
= 0; i
< delta_search_threads
; i
++) {
3003 if (!p
[i
].list_size
)
3005 pthread_mutex_init(&p
[i
].mutex
, NULL
);
3006 pthread_cond_init(&p
[i
].cond
, NULL
);
3007 ret
= pthread_create(&p
[i
].thread
, NULL
,
3008 threaded_find_deltas
, &p
[i
]);
3010 die(_("unable to create thread: %s"), strerror(ret
));
3015 * Now let's wait for work completion. Each time a thread is done
3016 * with its work, we steal half of the remaining work from the
3017 * thread with the largest number of unprocessed objects and give
3018 * it to that newly idle thread. This ensure good load balancing
3019 * until the remaining object list segments are simply too short
3020 * to be worth splitting anymore.
3022 while (active_threads
) {
3023 struct thread_params
*target
= NULL
;
3024 struct thread_params
*victim
= NULL
;
3025 unsigned sub_size
= 0;
3029 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
3034 pthread_cond_wait(&progress_cond
, &progress_mutex
);
3037 for (i
= 0; i
< delta_search_threads
; i
++)
3038 if (p
[i
].remaining
> 2*window
&&
3039 (!victim
|| victim
->remaining
< p
[i
].remaining
))
3042 sub_size
= victim
->remaining
/ 2;
3043 list
= victim
->list
+ victim
->list_size
- sub_size
;
3044 while (sub_size
&& list
[0]->hash
&&
3045 list
[0]->hash
== list
[-1]->hash
) {
3051 * It is possible for some "paths" to have
3052 * so many objects that no hash boundary
3053 * might be found. Let's just steal the
3054 * exact half in that case.
3056 sub_size
= victim
->remaining
/ 2;
3059 target
->list
= list
;
3060 victim
->list_size
-= sub_size
;
3061 victim
->remaining
-= sub_size
;
3063 target
->list_size
= sub_size
;
3064 target
->remaining
= sub_size
;
3065 target
->working
= 1;
3068 pthread_mutex_lock(&target
->mutex
);
3069 target
->data_ready
= 1;
3070 pthread_cond_signal(&target
->cond
);
3071 pthread_mutex_unlock(&target
->mutex
);
3074 pthread_join(target
->thread
, NULL
);
3075 pthread_cond_destroy(&target
->cond
);
3076 pthread_mutex_destroy(&target
->mutex
);
3080 cleanup_threaded_search();
3084 static int obj_is_packed(const struct object_id
*oid
)
3086 return packlist_find(&to_pack
, oid
) ||
3087 (reuse_packfile_bitmap
&&
3088 bitmap_walk_contains(bitmap_git
, reuse_packfile_bitmap
, oid
));
3091 static void add_tag_chain(const struct object_id
*oid
)
3096 * We catch duplicates already in add_object_entry(), but we'd
3097 * prefer to do this extra check to avoid having to parse the
3098 * tag at all if we already know that it's being packed (e.g., if
3099 * it was included via bitmaps, we would not have parsed it
3102 if (obj_is_packed(oid
))
3105 tag
= lookup_tag(the_repository
, oid
);
3107 if (!tag
|| parse_tag(tag
) || !tag
->tagged
)
3108 die(_("unable to pack objects reachable from tag %s"),
3111 add_object_entry(&tag
->object
.oid
, OBJ_TAG
, NULL
, 0);
3113 if (tag
->tagged
->type
!= OBJ_TAG
)
3116 tag
= (struct tag
*)tag
->tagged
;
3120 static int add_ref_tag(const char *tag UNUSED
, const struct object_id
*oid
,
3121 int flag UNUSED
, void *cb_data UNUSED
)
3123 struct object_id peeled
;
3125 if (!peel_iterated_oid(oid
, &peeled
) && obj_is_packed(&peeled
))
3130 static void prepare_pack(int window
, int depth
)
3132 struct object_entry
**delta_list
;
3133 uint32_t i
, nr_deltas
;
3136 if (use_delta_islands
)
3137 resolve_tree_islands(the_repository
, progress
, &to_pack
);
3139 get_object_details();
3142 * If we're locally repacking then we need to be doubly careful
3143 * from now on in order to make sure no stealth corruption gets
3144 * propagated to the new pack. Clients receiving streamed packs
3145 * should validate everything they get anyway so no need to incur
3146 * the additional cost here in that case.
3148 if (!pack_to_stdout
)
3149 do_check_packed_object_crc
= 1;
3151 if (!to_pack
.nr_objects
|| !window
|| !depth
)
3154 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
3157 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
3158 struct object_entry
*entry
= to_pack
.objects
+ i
;
3161 /* This happens if we decided to reuse existing
3162 * delta from a pack. "reuse_delta &&" is implied.
3166 if (!entry
->type_valid
||
3167 oe_size_less_than(&to_pack
, entry
, 50))
3170 if (entry
->no_try_delta
)
3173 if (!entry
->preferred_base
) {
3175 if (oe_type(entry
) < 0)
3176 die(_("unable to get type of object %s"),
3177 oid_to_hex(&entry
->idx
.oid
));
3179 if (oe_type(entry
) < 0) {
3181 * This object is not found, but we
3182 * don't have to include it anyway.
3188 delta_list
[n
++] = entry
;
3191 if (nr_deltas
&& n
> 1) {
3192 unsigned nr_done
= 0;
3195 progress_state
= start_progress(_("Compressing objects"),
3197 QSORT(delta_list
, n
, type_size_sort
);
3198 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
3199 stop_progress(&progress_state
);
3200 if (nr_done
!= nr_deltas
)
3201 die(_("inconsistency with delta count"));
3206 static int git_pack_config(const char *k
, const char *v
,
3207 const struct config_context
*ctx
, void *cb
)
3209 if (!strcmp(k
, "pack.window")) {
3210 window
= git_config_int(k
, v
, ctx
->kvi
);
3213 if (!strcmp(k
, "pack.windowmemory")) {
3214 window_memory_limit
= git_config_ulong(k
, v
, ctx
->kvi
);
3217 if (!strcmp(k
, "pack.depth")) {
3218 depth
= git_config_int(k
, v
, ctx
->kvi
);
3221 if (!strcmp(k
, "pack.deltacachesize")) {
3222 max_delta_cache_size
= git_config_int(k
, v
, ctx
->kvi
);
3225 if (!strcmp(k
, "pack.deltacachelimit")) {
3226 cache_max_small_delta_size
= git_config_int(k
, v
, ctx
->kvi
);
3229 if (!strcmp(k
, "pack.writebitmaphashcache")) {
3230 if (git_config_bool(k
, v
))
3231 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
3233 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
3236 if (!strcmp(k
, "pack.writebitmaplookuptable")) {
3237 if (git_config_bool(k
, v
))
3238 write_bitmap_options
|= BITMAP_OPT_LOOKUP_TABLE
;
3240 write_bitmap_options
&= ~BITMAP_OPT_LOOKUP_TABLE
;
3243 if (!strcmp(k
, "pack.usebitmaps")) {
3244 use_bitmap_index_default
= git_config_bool(k
, v
);
3247 if (!strcmp(k
, "pack.allowpackreuse")) {
3248 int res
= git_parse_maybe_bool_text(v
);
3250 if (!strcasecmp(v
, "single"))
3251 allow_pack_reuse
= SINGLE_PACK_REUSE
;
3252 else if (!strcasecmp(v
, "multi"))
3253 allow_pack_reuse
= MULTI_PACK_REUSE
;
3255 die(_("invalid pack.allowPackReuse value: '%s'"), v
);
3257 allow_pack_reuse
= SINGLE_PACK_REUSE
;
3259 allow_pack_reuse
= NO_PACK_REUSE
;
3263 if (!strcmp(k
, "pack.threads")) {
3264 delta_search_threads
= git_config_int(k
, v
, ctx
->kvi
);
3265 if (delta_search_threads
< 0)
3266 die(_("invalid number of threads specified (%d)"),
3267 delta_search_threads
);
3268 if (!HAVE_THREADS
&& delta_search_threads
!= 1) {
3269 warning(_("no threads support, ignoring %s"), k
);
3270 delta_search_threads
= 0;
3274 if (!strcmp(k
, "pack.indexversion")) {
3275 pack_idx_opts
.version
= git_config_int(k
, v
, ctx
->kvi
);
3276 if (pack_idx_opts
.version
> 2)
3277 die(_("bad pack.indexVersion=%"PRIu32
),
3278 pack_idx_opts
.version
);
3281 if (!strcmp(k
, "pack.writereverseindex")) {
3282 if (git_config_bool(k
, v
))
3283 pack_idx_opts
.flags
|= WRITE_REV
;
3285 pack_idx_opts
.flags
&= ~WRITE_REV
;
3288 if (!strcmp(k
, "uploadpack.blobpackfileuri")) {
3289 struct configured_exclusion
*ex
;
3290 const char *oid_end
, *pack_end
;
3292 * Stores the pack hash. This is not a true object ID, but is
3295 struct object_id pack_hash
;
3298 return config_error_nonbool(k
);
3300 ex
= xmalloc(sizeof(*ex
));
3301 if (parse_oid_hex(v
, &ex
->e
.oid
, &oid_end
) ||
3303 parse_oid_hex(oid_end
+ 1, &pack_hash
, &pack_end
) ||
3305 die(_("value of uploadpack.blobpackfileuri must be "
3306 "of the form '<object-hash> <pack-hash> <uri>' (got '%s')"), v
);
3307 if (oidmap_get(&configured_exclusions
, &ex
->e
.oid
))
3308 die(_("object already configured in another "
3309 "uploadpack.blobpackfileuri (got '%s')"), v
);
3310 ex
->pack_hash_hex
= xcalloc(1, pack_end
- oid_end
);
3311 memcpy(ex
->pack_hash_hex
, oid_end
+ 1, pack_end
- oid_end
- 1);
3312 ex
->uri
= xstrdup(pack_end
+ 1);
3313 oidmap_put(&configured_exclusions
, ex
);
3315 return git_default_config(k
, v
, ctx
, cb
);
3318 /* Counters for trace2 output when in --stdin-packs mode. */
3319 static int stdin_packs_found_nr
;
3320 static int stdin_packs_hints_nr
;
3322 static int add_object_entry_from_pack(const struct object_id
*oid
,
3323 struct packed_git
*p
,
3328 enum object_type type
= OBJ_NONE
;
3330 display_progress(progress_state
, ++nr_seen
);
3332 if (have_duplicate_entry(oid
, 0))
3335 ofs
= nth_packed_object_offset(p
, pos
);
3336 if (!want_object_in_pack(oid
, 0, &p
, &ofs
))
3340 struct rev_info
*revs
= _data
;
3341 struct object_info oi
= OBJECT_INFO_INIT
;
3344 if (packed_object_info(the_repository
, p
, ofs
, &oi
) < 0) {
3345 die(_("could not get type of object %s in pack %s"),
3346 oid_to_hex(oid
), p
->pack_name
);
3347 } else if (type
== OBJ_COMMIT
) {
3349 * commits in included packs are used as starting points for the
3350 * subsequent revision walk
3352 add_pending_oid(revs
, NULL
, oid
, 0);
3355 stdin_packs_found_nr
++;
3358 create_object_entry(oid
, type
, 0, 0, 0, p
, ofs
);
3363 static void show_commit_pack_hint(struct commit
*commit UNUSED
,
3366 /* nothing to do; commits don't have a namehash */
3369 static void show_object_pack_hint(struct object
*object
, const char *name
,
3372 struct object_entry
*oe
= packlist_find(&to_pack
, &object
->oid
);
3377 * Our 'to_pack' list was constructed by iterating all objects packed in
3378 * included packs, and so doesn't have a non-zero hash field that you
3379 * would typically pick up during a reachability traversal.
3381 * Make a best-effort attempt to fill in the ->hash and ->no_try_delta
3382 * here using a now in order to perhaps improve the delta selection
3385 oe
->hash
= pack_name_hash(name
);
3386 oe
->no_try_delta
= name
&& no_try_delta(name
);
3388 stdin_packs_hints_nr
++;
3391 static int pack_mtime_cmp(const void *_a
, const void *_b
)
3393 struct packed_git
*a
= ((const struct string_list_item
*)_a
)->util
;
3394 struct packed_git
*b
= ((const struct string_list_item
*)_b
)->util
;
3397 * order packs by descending mtime so that objects are laid out
3398 * roughly as newest-to-oldest
3400 if (a
->mtime
< b
->mtime
)
3402 else if (b
->mtime
< a
->mtime
)
3408 static void read_packs_list_from_stdin(void)
3410 struct strbuf buf
= STRBUF_INIT
;
3411 struct string_list include_packs
= STRING_LIST_INIT_DUP
;
3412 struct string_list exclude_packs
= STRING_LIST_INIT_DUP
;
3413 struct string_list_item
*item
= NULL
;
3415 struct packed_git
*p
;
3416 struct rev_info revs
;
3418 repo_init_revisions(the_repository
, &revs
, NULL
);
3420 * Use a revision walk to fill in the namehash of objects in the include
3421 * packs. To save time, we'll avoid traversing through objects that are
3422 * in excluded packs.
3424 * That may cause us to avoid populating all of the namehash fields of
3425 * all included objects, but our goal is best-effort, since this is only
3426 * an optimization during delta selection.
3428 revs
.no_kept_objects
= 1;
3429 revs
.keep_pack_cache_flags
|= IN_CORE_KEEP_PACKS
;
3430 revs
.blob_objects
= 1;
3431 revs
.tree_objects
= 1;
3432 revs
.tag_objects
= 1;
3433 revs
.ignore_missing_links
= 1;
3435 while (strbuf_getline(&buf
, stdin
) != EOF
) {
3439 if (*buf
.buf
== '^')
3440 string_list_append(&exclude_packs
, buf
.buf
+ 1);
3442 string_list_append(&include_packs
, buf
.buf
);
3447 string_list_sort(&include_packs
);
3448 string_list_remove_duplicates(&include_packs
, 0);
3449 string_list_sort(&exclude_packs
);
3450 string_list_remove_duplicates(&exclude_packs
, 0);
3452 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3453 const char *pack_name
= pack_basename(p
);
3455 if ((item
= string_list_lookup(&include_packs
, pack_name
)))
3457 if ((item
= string_list_lookup(&exclude_packs
, pack_name
)))
3462 * Arguments we got on stdin may not even be packs. First
3463 * check that to avoid segfaulting later on in
3464 * e.g. pack_mtime_cmp(), excluded packs are handled below.
3466 * Since we first parsed our STDIN and then sorted the input
3467 * lines the pack we error on will be whatever line happens to
3468 * sort first. This is lazy, it's enough that we report one
3469 * bad case here, we don't need to report the first/last one,
3472 for_each_string_list_item(item
, &include_packs
) {
3473 struct packed_git
*p
= item
->util
;
3475 die(_("could not find pack '%s'"), item
->string
);
3476 if (!is_pack_valid(p
))
3477 die(_("packfile %s cannot be accessed"), p
->pack_name
);
3481 * Then, handle all of the excluded packs, marking them as
3482 * kept in-core so that later calls to add_object_entry()
3483 * discards any objects that are also found in excluded packs.
3485 for_each_string_list_item(item
, &exclude_packs
) {
3486 struct packed_git
*p
= item
->util
;
3488 die(_("could not find pack '%s'"), item
->string
);
3489 p
->pack_keep_in_core
= 1;
3493 * Order packs by ascending mtime; use QSORT directly to access the
3494 * string_list_item's ->util pointer, which string_list_sort() does not
3497 QSORT(include_packs
.items
, include_packs
.nr
, pack_mtime_cmp
);
3499 for_each_string_list_item(item
, &include_packs
) {
3500 struct packed_git
*p
= item
->util
;
3501 for_each_object_in_pack(p
,
3502 add_object_entry_from_pack
,
3504 FOR_EACH_OBJECT_PACK_ORDER
);
3507 if (prepare_revision_walk(&revs
))
3508 die(_("revision walk setup failed"));
3509 traverse_commit_list(&revs
,
3510 show_commit_pack_hint
,
3511 show_object_pack_hint
,
3514 trace2_data_intmax("pack-objects", the_repository
, "stdin_packs_found",
3515 stdin_packs_found_nr
);
3516 trace2_data_intmax("pack-objects", the_repository
, "stdin_packs_hints",
3517 stdin_packs_hints_nr
);
3519 strbuf_release(&buf
);
3520 string_list_clear(&include_packs
, 0);
3521 string_list_clear(&exclude_packs
, 0);
3524 static void add_cruft_object_entry(const struct object_id
*oid
, enum object_type type
,
3525 struct packed_git
*pack
, off_t offset
,
3526 const char *name
, uint32_t mtime
)
3528 struct object_entry
*entry
;
3530 display_progress(progress_state
, ++nr_seen
);
3532 entry
= packlist_find(&to_pack
, oid
);
3535 entry
->hash
= pack_name_hash(name
);
3536 entry
->no_try_delta
= no_try_delta(name
);
3539 if (!want_object_in_pack(oid
, 0, &pack
, &offset
))
3541 if (!pack
&& type
== OBJ_BLOB
&& !has_loose_object(oid
)) {
3543 * If a traversed tree has a missing blob then we want
3544 * to avoid adding that missing object to our pack.
3546 * This only applies to missing blobs, not trees,
3547 * because the traversal needs to parse sub-trees but
3550 * Note we only perform this check when we couldn't
3551 * already find the object in a pack, so we're really
3552 * limited to "ensure non-tip blobs which don't exist in
3553 * packs do exist via loose objects". Confused?
3558 entry
= create_object_entry(oid
, type
, pack_name_hash(name
),
3559 0, name
&& no_try_delta(name
),
3563 if (mtime
> oe_cruft_mtime(&to_pack
, entry
))
3564 oe_set_cruft_mtime(&to_pack
, entry
, mtime
);
3568 static void show_cruft_object(struct object
*obj
, const char *name
, void *data UNUSED
)
3571 * if we did not record it earlier, it's at least as old as our
3572 * expiration value. Rather than find it exactly, just use that
3573 * value. This may bump it forward from its real mtime, but it
3574 * will still be "too old" next time we run with the same
3577 * if obj does appear in the packing list, this call is a noop (or may
3578 * set the namehash).
3580 add_cruft_object_entry(&obj
->oid
, obj
->type
, NULL
, 0, name
, cruft_expiration
);
3583 static void show_cruft_commit(struct commit
*commit
, void *data
)
3585 show_cruft_object((struct object
*)commit
, NULL
, data
);
3588 static int cruft_include_check_obj(struct object
*obj
, void *data UNUSED
)
3590 return !has_object_kept_pack(&obj
->oid
, IN_CORE_KEEP_PACKS
);
3593 static int cruft_include_check(struct commit
*commit
, void *data
)
3595 return cruft_include_check_obj((struct object
*)commit
, data
);
3598 static void set_cruft_mtime(const struct object
*object
,
3599 struct packed_git
*pack
,
3600 off_t offset
, time_t mtime
)
3602 add_cruft_object_entry(&object
->oid
, object
->type
, pack
, offset
, NULL
,
3606 static void mark_pack_kept_in_core(struct string_list
*packs
, unsigned keep
)
3608 struct string_list_item
*item
= NULL
;
3609 for_each_string_list_item(item
, packs
) {
3610 struct packed_git
*p
= item
->util
;
3612 die(_("could not find pack '%s'"), item
->string
);
3613 p
->pack_keep_in_core
= keep
;
3617 static void add_unreachable_loose_objects(void);
3618 static void add_objects_in_unpacked_packs(void);
3620 static void enumerate_cruft_objects(void)
3623 progress_state
= start_progress(_("Enumerating cruft objects"), 0);
3625 add_objects_in_unpacked_packs();
3626 add_unreachable_loose_objects();
3628 stop_progress(&progress_state
);
3631 static void enumerate_and_traverse_cruft_objects(struct string_list
*fresh_packs
)
3633 struct packed_git
*p
;
3634 struct rev_info revs
;
3637 repo_init_revisions(the_repository
, &revs
, NULL
);
3639 revs
.tag_objects
= 1;
3640 revs
.tree_objects
= 1;
3641 revs
.blob_objects
= 1;
3643 revs
.include_check
= cruft_include_check
;
3644 revs
.include_check_obj
= cruft_include_check_obj
;
3646 revs
.ignore_missing_links
= 1;
3649 progress_state
= start_progress(_("Enumerating cruft objects"), 0);
3650 ret
= add_unseen_recent_objects_to_traversal(&revs
, cruft_expiration
,
3651 set_cruft_mtime
, 1);
3652 stop_progress(&progress_state
);
3655 die(_("unable to add cruft objects"));
3658 * Re-mark only the fresh packs as kept so that objects in
3659 * unknown packs do not halt the reachability traversal early.
3661 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
)
3662 p
->pack_keep_in_core
= 0;
3663 mark_pack_kept_in_core(fresh_packs
, 1);
3665 if (prepare_revision_walk(&revs
))
3666 die(_("revision walk setup failed"));
3668 progress_state
= start_progress(_("Traversing cruft objects"), 0);
3670 traverse_commit_list(&revs
, show_cruft_commit
, show_cruft_object
, NULL
);
3672 stop_progress(&progress_state
);
3675 static void read_cruft_objects(void)
3677 struct strbuf buf
= STRBUF_INIT
;
3678 struct string_list discard_packs
= STRING_LIST_INIT_DUP
;
3679 struct string_list fresh_packs
= STRING_LIST_INIT_DUP
;
3680 struct packed_git
*p
;
3682 ignore_packed_keep_in_core
= 1;
3684 while (strbuf_getline(&buf
, stdin
) != EOF
) {
3688 if (*buf
.buf
== '-')
3689 string_list_append(&discard_packs
, buf
.buf
+ 1);
3691 string_list_append(&fresh_packs
, buf
.buf
);
3694 string_list_sort(&discard_packs
);
3695 string_list_sort(&fresh_packs
);
3697 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3698 const char *pack_name
= pack_basename(p
);
3699 struct string_list_item
*item
;
3701 item
= string_list_lookup(&fresh_packs
, pack_name
);
3703 item
= string_list_lookup(&discard_packs
, pack_name
);
3709 * This pack wasn't mentioned in either the "fresh" or
3710 * "discard" list, so the caller didn't know about it.
3712 * Mark it as kept so that its objects are ignored by
3713 * add_unseen_recent_objects_to_traversal(). We'll
3714 * unmark it before starting the traversal so it doesn't
3715 * halt the traversal early.
3717 p
->pack_keep_in_core
= 1;
3721 mark_pack_kept_in_core(&fresh_packs
, 1);
3722 mark_pack_kept_in_core(&discard_packs
, 0);
3724 if (cruft_expiration
)
3725 enumerate_and_traverse_cruft_objects(&fresh_packs
);
3727 enumerate_cruft_objects();
3729 strbuf_release(&buf
);
3730 string_list_clear(&discard_packs
, 0);
3731 string_list_clear(&fresh_packs
, 0);
3734 static void read_object_list_from_stdin(void)
3736 char line
[GIT_MAX_HEXSZ
+ 1 + PATH_MAX
+ 2];
3737 struct object_id oid
;
3741 if (!fgets(line
, sizeof(line
), stdin
)) {
3745 BUG("fgets returned NULL, not EOF, not error!");
3751 if (line
[0] == '-') {
3752 if (get_oid_hex(line
+1, &oid
))
3753 die(_("expected edge object ID, got garbage:\n %s"),
3755 add_preferred_base(&oid
);
3758 if (parse_oid_hex(line
, &oid
, &p
))
3759 die(_("expected object ID, got garbage:\n %s"), line
);
3761 add_preferred_base_object(p
+ 1);
3762 add_object_entry(&oid
, OBJ_NONE
, p
+ 1, 0);
3766 static void show_commit(struct commit
*commit
, void *data UNUSED
)
3768 add_object_entry(&commit
->object
.oid
, OBJ_COMMIT
, NULL
, 0);
3770 if (write_bitmap_index
)
3771 index_commit_for_bitmap(commit
);
3773 if (use_delta_islands
)
3774 propagate_island_marks(commit
);
3777 static void show_object(struct object
*obj
, const char *name
,
3780 add_preferred_base_object(name
);
3781 add_object_entry(&obj
->oid
, obj
->type
, name
, 0);
3783 if (use_delta_islands
) {
3786 struct object_entry
*ent
;
3788 /* the empty string is a root tree, which is depth 0 */
3789 depth
= *name
? 1 : 0;
3790 for (p
= strchr(name
, '/'); p
; p
= strchr(p
+ 1, '/'))
3793 ent
= packlist_find(&to_pack
, &obj
->oid
);
3794 if (ent
&& depth
> oe_tree_depth(&to_pack
, ent
))
3795 oe_set_tree_depth(&to_pack
, ent
, depth
);
3799 static void show_object__ma_allow_any(struct object
*obj
, const char *name
, void *data
)
3801 assert(arg_missing_action
== MA_ALLOW_ANY
);
3804 * Quietly ignore ALL missing objects. This avoids problems with
3805 * staging them now and getting an odd error later.
3807 if (!has_object(the_repository
, &obj
->oid
, 0))
3810 show_object(obj
, name
, data
);
3813 static void show_object__ma_allow_promisor(struct object
*obj
, const char *name
, void *data
)
3815 assert(arg_missing_action
== MA_ALLOW_PROMISOR
);
3818 * Quietly ignore EXPECTED missing objects. This avoids problems with
3819 * staging them now and getting an odd error later.
3821 if (!has_object(the_repository
, &obj
->oid
, 0) && is_promisor_object(&obj
->oid
))
3824 show_object(obj
, name
, data
);
3827 static int option_parse_missing_action(const struct option
*opt UNUSED
,
3828 const char *arg
, int unset
)
3833 if (!strcmp(arg
, "error")) {
3834 arg_missing_action
= MA_ERROR
;
3835 fn_show_object
= show_object
;
3839 if (!strcmp(arg
, "allow-any")) {
3840 arg_missing_action
= MA_ALLOW_ANY
;
3841 fetch_if_missing
= 0;
3842 fn_show_object
= show_object__ma_allow_any
;
3846 if (!strcmp(arg
, "allow-promisor")) {
3847 arg_missing_action
= MA_ALLOW_PROMISOR
;
3848 fetch_if_missing
= 0;
3849 fn_show_object
= show_object__ma_allow_promisor
;
3853 die(_("invalid value for '%s': '%s'"), "--missing", arg
);
3857 static void show_edge(struct commit
*commit
)
3859 add_preferred_base(&commit
->object
.oid
);
3862 static int add_object_in_unpacked_pack(const struct object_id
*oid
,
3863 struct packed_git
*pack
,
3871 if (pack
->is_cruft
) {
3872 if (load_pack_mtimes(pack
) < 0)
3873 die(_("could not load cruft pack .mtimes"));
3874 mtime
= nth_packed_mtime(pack
, pos
);
3876 mtime
= pack
->mtime
;
3878 offset
= nth_packed_object_offset(pack
, pos
);
3880 add_cruft_object_entry(oid
, OBJ_NONE
, pack
, offset
,
3883 add_object_entry(oid
, OBJ_NONE
, "", 0);
3888 static void add_objects_in_unpacked_packs(void)
3890 if (for_each_packed_object(add_object_in_unpacked_pack
, NULL
,
3891 FOR_EACH_OBJECT_PACK_ORDER
|
3892 FOR_EACH_OBJECT_LOCAL_ONLY
|
3893 FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS
|
3894 FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS
))
3895 die(_("cannot open pack index"));
3898 static int add_loose_object(const struct object_id
*oid
, const char *path
,
3901 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
3904 warning(_("loose object at %s could not be examined"), path
);
3910 if (stat(path
, &st
) < 0) {
3911 if (errno
== ENOENT
)
3913 return error_errno("unable to stat %s", oid_to_hex(oid
));
3916 add_cruft_object_entry(oid
, type
, NULL
, 0, NULL
,
3919 add_object_entry(oid
, type
, "", 0);
3925 * We actually don't even have to worry about reachability here.
3926 * add_object_entry will weed out duplicates, so we just add every
3927 * loose object we find.
3929 static void add_unreachable_loose_objects(void)
3931 for_each_loose_file_in_objdir(get_object_directory(),
3936 static int has_sha1_pack_kept_or_nonlocal(const struct object_id
*oid
)
3938 static struct packed_git
*last_found
= (void *)1;
3939 struct packed_git
*p
;
3941 p
= (last_found
!= (void *)1) ? last_found
:
3942 get_all_packs(the_repository
);
3945 if ((!p
->pack_local
|| p
->pack_keep
||
3946 p
->pack_keep_in_core
) &&
3947 find_pack_entry_one(oid
->hash
, p
)) {
3951 if (p
== last_found
)
3952 p
= get_all_packs(the_repository
);
3955 if (p
== last_found
)
3962 * Store a list of sha1s that are should not be discarded
3963 * because they are either written too recently, or are
3964 * reachable from another object that was.
3966 * This is filled by get_object_list.
3968 static struct oid_array recent_objects
;
3970 static int loosened_object_can_be_discarded(const struct object_id
*oid
,
3973 if (!unpack_unreachable_expiration
)
3975 if (mtime
> unpack_unreachable_expiration
)
3977 if (oid_array_lookup(&recent_objects
, oid
) >= 0)
3982 static void loosen_unused_packed_objects(void)
3984 struct packed_git
*p
;
3986 uint32_t loosened_objects_nr
= 0;
3987 struct object_id oid
;
3989 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3990 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
3993 if (open_pack_index(p
))
3994 die(_("cannot open pack index"));
3996 for (i
= 0; i
< p
->num_objects
; i
++) {
3997 nth_packed_object_id(&oid
, p
, i
);
3998 if (!packlist_find(&to_pack
, &oid
) &&
3999 !has_sha1_pack_kept_or_nonlocal(&oid
) &&
4000 !loosened_object_can_be_discarded(&oid
, p
->mtime
)) {
4001 if (force_object_loose(&oid
, p
->mtime
))
4002 die(_("unable to force loose object"));
4003 loosened_objects_nr
++;
4008 trace2_data_intmax("pack-objects", the_repository
,
4009 "loosen_unused_packed_objects/loosened", loosened_objects_nr
);
4013 * This tracks any options which pack-reuse code expects to be on, or which a
4014 * reader of the pack might not understand, and which would therefore prevent
4015 * blind reuse of what we have on disk.
4017 static int pack_options_allow_reuse(void)
4019 return allow_pack_reuse
!= NO_PACK_REUSE
&&
4021 !ignore_packed_keep_on_disk
&&
4022 !ignore_packed_keep_in_core
&&
4023 (!local
|| !have_non_local_packs
) &&
4027 static int get_object_list_from_bitmap(struct rev_info
*revs
)
4029 if (!(bitmap_git
= prepare_bitmap_walk(revs
, 0)))
4032 if (pack_options_allow_reuse())
4033 reuse_partial_packfile_from_bitmap(bitmap_git
,
4035 &reuse_packfiles_nr
,
4036 &reuse_packfile_bitmap
,
4037 allow_pack_reuse
== MULTI_PACK_REUSE
);
4039 if (reuse_packfiles
) {
4040 reuse_packfile_objects
= bitmap_popcount(reuse_packfile_bitmap
);
4041 if (!reuse_packfile_objects
)
4042 BUG("expected non-empty reuse bitmap");
4044 nr_result
+= reuse_packfile_objects
;
4045 nr_seen
+= reuse_packfile_objects
;
4046 display_progress(progress_state
, nr_seen
);
4049 traverse_bitmap_commit_list(bitmap_git
, revs
,
4050 &add_object_entry_from_bitmap
);
4054 static void record_recent_object(struct object
*obj
,
4055 const char *name UNUSED
,
4058 oid_array_append(&recent_objects
, &obj
->oid
);
4061 static void record_recent_commit(struct commit
*commit
, void *data UNUSED
)
4063 oid_array_append(&recent_objects
, &commit
->object
.oid
);
4066 static int mark_bitmap_preferred_tip(const char *refname
,
4067 const struct object_id
*oid
,
4071 struct object_id peeled
;
4072 struct object
*object
;
4074 if (!peel_iterated_oid(oid
, &peeled
))
4077 object
= parse_object_or_die(oid
, refname
);
4078 if (object
->type
== OBJ_COMMIT
)
4079 object
->flags
|= NEEDS_BITMAP
;
4084 static void mark_bitmap_preferred_tips(void)
4086 struct string_list_item
*item
;
4087 const struct string_list
*preferred_tips
;
4089 preferred_tips
= bitmap_preferred_tips(the_repository
);
4090 if (!preferred_tips
)
4093 for_each_string_list_item(item
, preferred_tips
) {
4094 for_each_ref_in(item
->string
, mark_bitmap_preferred_tip
, NULL
);
4098 static void get_object_list(struct rev_info
*revs
, int ac
, const char **av
)
4100 struct setup_revision_opt s_r_opt
= {
4101 .allow_exclude_promisor_objects
= 1,
4107 save_commit_buffer
= 0;
4108 setup_revisions(ac
, av
, revs
, &s_r_opt
);
4110 /* make sure shallows are read */
4111 is_repository_shallow(the_repository
);
4113 save_warning
= warn_on_object_refname_ambiguity
;
4114 warn_on_object_refname_ambiguity
= 0;
4116 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
4117 int len
= strlen(line
);
4118 if (len
&& line
[len
- 1] == '\n')
4123 if (!strcmp(line
, "--not")) {
4124 flags
^= UNINTERESTING
;
4125 write_bitmap_index
= 0;
4128 if (starts_with(line
, "--shallow ")) {
4129 struct object_id oid
;
4130 if (get_oid_hex(line
+ 10, &oid
))
4131 die("not an object name '%s'", line
+ 10);
4132 register_shallow(the_repository
, &oid
);
4133 use_bitmap_index
= 0;
4136 die(_("not a rev '%s'"), line
);
4138 if (handle_revision_arg(line
, revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
4139 die(_("bad revision '%s'"), line
);
4142 warn_on_object_refname_ambiguity
= save_warning
;
4144 if (use_bitmap_index
&& !get_object_list_from_bitmap(revs
))
4147 if (use_delta_islands
)
4148 load_delta_islands(the_repository
, progress
);
4150 if (write_bitmap_index
)
4151 mark_bitmap_preferred_tips();
4153 if (prepare_revision_walk(revs
))
4154 die(_("revision walk setup failed"));
4155 mark_edges_uninteresting(revs
, show_edge
, sparse
);
4157 if (!fn_show_object
)
4158 fn_show_object
= show_object
;
4159 traverse_commit_list(revs
,
4160 show_commit
, fn_show_object
,
4163 if (unpack_unreachable_expiration
) {
4164 revs
->ignore_missing_links
= 1;
4165 if (add_unseen_recent_objects_to_traversal(revs
,
4166 unpack_unreachable_expiration
, NULL
, 0))
4167 die(_("unable to add recent objects"));
4168 if (prepare_revision_walk(revs
))
4169 die(_("revision walk setup failed"));
4170 traverse_commit_list(revs
, record_recent_commit
,
4171 record_recent_object
, NULL
);
4174 if (keep_unreachable
)
4175 add_objects_in_unpacked_packs();
4176 if (pack_loose_unreachable
)
4177 add_unreachable_loose_objects();
4178 if (unpack_unreachable
)
4179 loosen_unused_packed_objects();
4181 oid_array_clear(&recent_objects
);
4184 static void add_extra_kept_packs(const struct string_list
*names
)
4186 struct packed_git
*p
;
4191 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
4192 const char *name
= basename(p
->pack_name
);
4198 for (i
= 0; i
< names
->nr
; i
++)
4199 if (!fspathcmp(name
, names
->items
[i
].string
))
4202 if (i
< names
->nr
) {
4203 p
->pack_keep_in_core
= 1;
4204 ignore_packed_keep_in_core
= 1;
4210 static int option_parse_quiet(const struct option
*opt
, const char *arg
,
4213 int *val
= opt
->value
;
4215 BUG_ON_OPT_ARG(arg
);
4224 static int option_parse_index_version(const struct option
*opt
,
4225 const char *arg
, int unset
)
4227 struct pack_idx_option
*popts
= opt
->value
;
4229 const char *val
= arg
;
4231 BUG_ON_OPT_NEG(unset
);
4233 popts
->version
= strtoul(val
, &c
, 10);
4234 if (popts
->version
> 2)
4235 die(_("unsupported index version %s"), val
);
4236 if (*c
== ',' && c
[1])
4237 popts
->off32_limit
= strtoul(c
+1, &c
, 0);
4238 if (*c
|| popts
->off32_limit
& 0x80000000)
4239 die(_("bad index version '%s'"), val
);
4243 static int option_parse_unpack_unreachable(const struct option
*opt UNUSED
,
4244 const char *arg
, int unset
)
4247 unpack_unreachable
= 0;
4248 unpack_unreachable_expiration
= 0;
4251 unpack_unreachable
= 1;
4253 unpack_unreachable_expiration
= approxidate(arg
);
4258 static int option_parse_cruft_expiration(const struct option
*opt UNUSED
,
4259 const char *arg
, int unset
)
4263 cruft_expiration
= 0;
4267 cruft_expiration
= approxidate(arg
);
4272 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
4274 int use_internal_rev_list
= 0;
4276 int all_progress_implied
= 0;
4277 struct strvec rp
= STRVEC_INIT
;
4278 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
4279 int rev_list_index
= 0;
4280 int stdin_packs
= 0;
4281 struct string_list keep_pack_list
= STRING_LIST_INIT_NODUP
;
4282 struct list_objects_filter_options filter_options
=
4283 LIST_OBJECTS_FILTER_INIT
;
4285 struct option pack_objects_options
[] = {
4286 OPT_CALLBACK_F('q', "quiet", &progress
, NULL
,
4287 N_("do not show progress meter"),
4288 PARSE_OPT_NOARG
, option_parse_quiet
),
4289 OPT_SET_INT(0, "progress", &progress
,
4290 N_("show progress meter"), 1),
4291 OPT_SET_INT(0, "all-progress", &progress
,
4292 N_("show progress meter during object writing phase"), 2),
4293 OPT_BOOL(0, "all-progress-implied",
4294 &all_progress_implied
,
4295 N_("similar to --all-progress when progress meter is shown")),
4296 OPT_CALLBACK_F(0, "index-version", &pack_idx_opts
, N_("<version>[,<offset>]"),
4297 N_("write the pack index file in the specified idx format version"),
4298 PARSE_OPT_NONEG
, option_parse_index_version
),
4299 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
4300 N_("maximum size of each output pack file")),
4301 OPT_BOOL(0, "local", &local
,
4302 N_("ignore borrowed objects from alternate object store")),
4303 OPT_BOOL(0, "incremental", &incremental
,
4304 N_("ignore packed objects")),
4305 OPT_INTEGER(0, "window", &window
,
4306 N_("limit pack window by objects")),
4307 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
4308 N_("limit pack window by memory in addition to object limit")),
4309 OPT_INTEGER(0, "depth", &depth
,
4310 N_("maximum length of delta chain allowed in the resulting pack")),
4311 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
4312 N_("reuse existing deltas")),
4313 OPT_BOOL(0, "reuse-object", &reuse_object
,
4314 N_("reuse existing objects")),
4315 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
4316 N_("use OFS_DELTA objects")),
4317 OPT_INTEGER(0, "threads", &delta_search_threads
,
4318 N_("use threads when searching for best delta matches")),
4319 OPT_BOOL(0, "non-empty", &non_empty
,
4320 N_("do not create an empty pack output")),
4321 OPT_BOOL(0, "revs", &use_internal_rev_list
,
4322 N_("read revision arguments from standard input")),
4323 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked
,
4324 N_("limit the objects to those that are not yet packed"),
4325 1, PARSE_OPT_NONEG
),
4326 OPT_SET_INT_F(0, "all", &rev_list_all
,
4327 N_("include objects reachable from any reference"),
4328 1, PARSE_OPT_NONEG
),
4329 OPT_SET_INT_F(0, "reflog", &rev_list_reflog
,
4330 N_("include objects referred by reflog entries"),
4331 1, PARSE_OPT_NONEG
),
4332 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index
,
4333 N_("include objects referred to by the index"),
4334 1, PARSE_OPT_NONEG
),
4335 OPT_BOOL(0, "stdin-packs", &stdin_packs
,
4336 N_("read packs from stdin")),
4337 OPT_BOOL(0, "stdout", &pack_to_stdout
,
4338 N_("output pack to stdout")),
4339 OPT_BOOL(0, "include-tag", &include_tag
,
4340 N_("include tag objects that refer to objects to be packed")),
4341 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
4342 N_("keep unreachable objects")),
4343 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
4344 N_("pack loose unreachable objects")),
4345 OPT_CALLBACK_F(0, "unpack-unreachable", NULL
, N_("time"),
4346 N_("unpack unreachable objects newer than <time>"),
4347 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
),
4348 OPT_BOOL(0, "cruft", &cruft
, N_("create a cruft pack")),
4349 OPT_CALLBACK_F(0, "cruft-expiration", NULL
, N_("time"),
4350 N_("expire cruft objects older than <time>"),
4351 PARSE_OPT_OPTARG
, option_parse_cruft_expiration
),
4352 OPT_BOOL(0, "sparse", &sparse
,
4353 N_("use the sparse reachability algorithm")),
4354 OPT_BOOL(0, "thin", &thin
,
4355 N_("create thin packs")),
4356 OPT_BOOL(0, "shallow", &shallow
,
4357 N_("create packs suitable for shallow fetches")),
4358 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk
,
4359 N_("ignore packs that have companion .keep file")),
4360 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list
, N_("name"),
4361 N_("ignore this pack")),
4362 OPT_INTEGER(0, "compression", &pack_compression_level
,
4363 N_("pack compression level")),
4364 OPT_BOOL(0, "keep-true-parents", &grafts_keep_true_parents
,
4365 N_("do not hide commits by grafts")),
4366 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
4367 N_("use a bitmap index if available to speed up counting objects")),
4368 OPT_SET_INT(0, "write-bitmap-index", &write_bitmap_index
,
4369 N_("write a bitmap index together with the pack index"),
4371 OPT_SET_INT_F(0, "write-bitmap-index-quiet",
4372 &write_bitmap_index
,
4373 N_("write a bitmap index if possible"),
4374 WRITE_BITMAP_QUIET
, PARSE_OPT_HIDDEN
),
4375 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options
),
4376 OPT_CALLBACK_F(0, "missing", NULL
, N_("action"),
4377 N_("handling for missing objects"), PARSE_OPT_NONEG
,
4378 option_parse_missing_action
),
4379 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects
,
4380 N_("do not pack objects in promisor packfiles")),
4381 OPT_BOOL(0, "delta-islands", &use_delta_islands
,
4382 N_("respect islands during delta compression")),
4383 OPT_STRING_LIST(0, "uri-protocol", &uri_protocols
,
4385 N_("exclude any configured uploadpack.blobpackfileuri with this protocol")),
4389 if (DFS_NUM_STATES
> (1 << OE_DFS_STATE_BITS
))
4390 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
4392 disable_replace_refs();
4394 sparse
= git_env_bool("GIT_TEST_PACK_SPARSE", -1);
4395 if (the_repository
->gitdir
) {
4396 prepare_repo_settings(the_repository
);
4398 sparse
= the_repository
->settings
.pack_use_sparse
;
4401 reset_pack_idx_option(&pack_idx_opts
);
4402 pack_idx_opts
.flags
|= WRITE_REV
;
4403 git_config(git_pack_config
, NULL
);
4404 if (git_env_bool(GIT_TEST_NO_WRITE_REV_INDEX
, 0))
4405 pack_idx_opts
.flags
&= ~WRITE_REV
;
4407 progress
= isatty(2);
4408 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
4412 base_name
= argv
[0];
4415 if (pack_to_stdout
!= !base_name
|| argc
)
4416 usage_with_options(pack_usage
, pack_objects_options
);
4420 if (depth
>= (1 << OE_DEPTH_BITS
)) {
4421 warning(_("delta chain depth %d is too deep, forcing %d"),
4422 depth
, (1 << OE_DEPTH_BITS
) - 1);
4423 depth
= (1 << OE_DEPTH_BITS
) - 1;
4425 if (cache_max_small_delta_size
>= (1U << OE_Z_DELTA_BITS
)) {
4426 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
4427 (1U << OE_Z_DELTA_BITS
) - 1);
4428 cache_max_small_delta_size
= (1U << OE_Z_DELTA_BITS
) - 1;
4433 strvec_push(&rp
, "pack-objects");
4435 use_internal_rev_list
= 1;
4436 strvec_push(&rp
, shallow
4437 ? "--objects-edge-aggressive"
4438 : "--objects-edge");
4440 strvec_push(&rp
, "--objects");
4443 use_internal_rev_list
= 1;
4444 strvec_push(&rp
, "--all");
4446 if (rev_list_reflog
) {
4447 use_internal_rev_list
= 1;
4448 strvec_push(&rp
, "--reflog");
4450 if (rev_list_index
) {
4451 use_internal_rev_list
= 1;
4452 strvec_push(&rp
, "--indexed-objects");
4454 if (rev_list_unpacked
&& !stdin_packs
) {
4455 use_internal_rev_list
= 1;
4456 strvec_push(&rp
, "--unpacked");
4459 if (exclude_promisor_objects
) {
4460 use_internal_rev_list
= 1;
4461 fetch_if_missing
= 0;
4462 strvec_push(&rp
, "--exclude-promisor-objects");
4464 if (unpack_unreachable
|| keep_unreachable
|| pack_loose_unreachable
)
4465 use_internal_rev_list
= 1;
4469 if (pack_compression_level
== -1)
4470 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
4471 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
4472 die(_("bad pack compression level %d"), pack_compression_level
);
4474 if (!delta_search_threads
) /* --threads=0 means autodetect */
4475 delta_search_threads
= online_cpus();
4477 if (!HAVE_THREADS
&& delta_search_threads
!= 1)
4478 warning(_("no threads support, ignoring --threads"));
4479 if (!pack_to_stdout
&& !pack_size_limit
)
4480 pack_size_limit
= pack_size_limit_cfg
;
4481 if (pack_to_stdout
&& pack_size_limit
)
4482 die(_("--max-pack-size cannot be used to build a pack for transfer"));
4483 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
4484 warning(_("minimum pack size limit is 1 MiB"));
4485 pack_size_limit
= 1024*1024;
4488 if (!pack_to_stdout
&& thin
)
4489 die(_("--thin cannot be used to build an indexable pack"));
4491 if (keep_unreachable
&& unpack_unreachable
)
4492 die(_("options '%s' and '%s' cannot be used together"), "--keep-unreachable", "--unpack-unreachable");
4493 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
4494 unpack_unreachable_expiration
= 0;
4496 if (stdin_packs
&& filter_options
.choice
)
4497 die(_("cannot use --filter with --stdin-packs"));
4499 if (stdin_packs
&& use_internal_rev_list
)
4500 die(_("cannot use internal rev list with --stdin-packs"));
4503 if (use_internal_rev_list
)
4504 die(_("cannot use internal rev list with --cruft"));
4506 die(_("cannot use --stdin-packs with --cruft"));
4510 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
4512 * - to produce good pack (with bitmap index not-yet-packed objects are
4513 * packed in suboptimal order).
4515 * - to use more robust pack-generation codepath (avoiding possible
4516 * bugs in bitmap code and possible bitmap index corruption).
4518 if (!pack_to_stdout
)
4519 use_bitmap_index_default
= 0;
4521 if (use_bitmap_index
< 0)
4522 use_bitmap_index
= use_bitmap_index_default
;
4524 /* "hard" reasons not to use bitmaps; these just won't work at all */
4525 if (!use_internal_rev_list
|| (!pack_to_stdout
&& write_bitmap_index
) || is_repository_shallow(the_repository
))
4526 use_bitmap_index
= 0;
4528 if (pack_to_stdout
|| !rev_list_all
)
4529 write_bitmap_index
= 0;
4531 if (use_delta_islands
)
4532 strvec_push(&rp
, "--topo-order");
4534 if (progress
&& all_progress_implied
)
4537 add_extra_kept_packs(&keep_pack_list
);
4538 if (ignore_packed_keep_on_disk
) {
4539 struct packed_git
*p
;
4540 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
)
4541 if (p
->pack_local
&& p
->pack_keep
)
4543 if (!p
) /* no keep-able packs found */
4544 ignore_packed_keep_on_disk
= 0;
4548 * unlike ignore_packed_keep_on_disk above, we do not
4549 * want to unset "local" based on looking at packs, as
4550 * it also covers non-local objects
4552 struct packed_git
*p
;
4553 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
4554 if (!p
->pack_local
) {
4555 have_non_local_packs
= 1;
4561 trace2_region_enter("pack-objects", "enumerate-objects",
4563 prepare_packing_data(the_repository
, &to_pack
);
4565 if (progress
&& !cruft
)
4566 progress_state
= start_progress(_("Enumerating objects"), 0);
4568 /* avoids adding objects in excluded packs */
4569 ignore_packed_keep_in_core
= 1;
4570 read_packs_list_from_stdin();
4571 if (rev_list_unpacked
)
4572 add_unreachable_loose_objects();
4574 read_cruft_objects();
4575 } else if (!use_internal_rev_list
) {
4576 read_object_list_from_stdin();
4578 struct rev_info revs
;
4580 repo_init_revisions(the_repository
, &revs
, NULL
);
4581 list_objects_filter_copy(&revs
.filter
, &filter_options
);
4582 get_object_list(&revs
, rp
.nr
, rp
.v
);
4583 release_revisions(&revs
);
4585 cleanup_preferred_base();
4586 if (include_tag
&& nr_result
)
4587 for_each_tag_ref(add_ref_tag
, NULL
);
4588 stop_progress(&progress_state
);
4589 trace2_region_leave("pack-objects", "enumerate-objects",
4592 if (non_empty
&& !nr_result
)
4595 trace2_region_enter("pack-objects", "prepare-pack",
4597 prepare_pack(window
, depth
);
4598 trace2_region_leave("pack-objects", "prepare-pack",
4602 trace2_region_enter("pack-objects", "write-pack-file", the_repository
);
4603 write_excluded_by_configs();
4605 trace2_region_leave("pack-objects", "write-pack-file", the_repository
);
4609 _("Total %"PRIu32
" (delta %"PRIu32
"),"
4610 " reused %"PRIu32
" (delta %"PRIu32
"),"
4611 " pack-reused %"PRIu32
" (from %"PRIuMAX
")"),
4612 written
, written_delta
, reused
, reused_delta
,
4613 reuse_packfile_objects
,
4614 (uintmax_t)reuse_packfiles_used_nr
);
4616 trace2_data_intmax("pack-objects", the_repository
, "written", written
);
4617 trace2_data_intmax("pack-objects", the_repository
, "written/delta", written_delta
);
4618 trace2_data_intmax("pack-objects", the_repository
, "reused", reused
);
4619 trace2_data_intmax("pack-objects", the_repository
, "reused/delta", reused_delta
);
4620 trace2_data_intmax("pack-objects", the_repository
, "pack-reused", reuse_packfile_objects
);
4621 trace2_data_intmax("pack-objects", the_repository
, "packs-reused", reuse_packfiles_used_nr
);
4624 clear_packing_data(&to_pack
);
4625 list_objects_filter_release(&filter_options
);