5 #include "repository.h"
15 #include "pack-revindex.h"
16 #include "csum-file.h"
17 #include "tree-walk.h"
20 #include "list-objects.h"
21 #include "list-objects-filter.h"
22 #include "list-objects-filter-options.h"
23 #include "pack-objects.h"
26 #include "streaming.h"
27 #include "thread-utils.h"
28 #include "pack-bitmap.h"
29 #include "delta-islands.h"
30 #include "reachable.h"
31 #include "oid-array.h"
35 #include "object-store.h"
36 #include "replace-object.h"
41 #include "promisor-remote.h"
42 #include "pack-mtimes.h"
45 * Objects we are going to pack are collected in the `to_pack` structure.
46 * It contains an array (dynamically expanded) of the object data, and a map
47 * that can resolve SHA1s to their position in the array.
49 static struct packing_data to_pack
;
51 static inline struct object_entry
*oe_delta(
52 const struct packing_data
*pack
,
53 const struct object_entry
*e
)
58 return &pack
->ext_bases
[e
->delta_idx
- 1];
60 return &pack
->objects
[e
->delta_idx
- 1];
63 static inline unsigned long oe_delta_size(struct packing_data
*pack
,
64 const struct object_entry
*e
)
66 if (e
->delta_size_valid
)
67 return e
->delta_size_
;
70 * pack->delta_size[] can't be NULL because oe_set_delta_size()
71 * must have been called when a new delta is saved with
73 * If oe_delta() returns NULL (i.e. default state, which means
74 * delta_size_valid is also false), then the caller must never
75 * call oe_delta_size().
77 return pack
->delta_size
[e
- pack
->objects
];
80 unsigned long oe_get_size_slow(struct packing_data
*pack
,
81 const struct object_entry
*e
);
83 static inline unsigned long oe_size(struct packing_data
*pack
,
84 const struct object_entry
*e
)
89 return oe_get_size_slow(pack
, e
);
92 static inline void oe_set_delta(struct packing_data
*pack
,
93 struct object_entry
*e
,
94 struct object_entry
*delta
)
97 e
->delta_idx
= (delta
- pack
->objects
) + 1;
102 static inline struct object_entry
*oe_delta_sibling(
103 const struct packing_data
*pack
,
104 const struct object_entry
*e
)
106 if (e
->delta_sibling_idx
)
107 return &pack
->objects
[e
->delta_sibling_idx
- 1];
111 static inline struct object_entry
*oe_delta_child(
112 const struct packing_data
*pack
,
113 const struct object_entry
*e
)
115 if (e
->delta_child_idx
)
116 return &pack
->objects
[e
->delta_child_idx
- 1];
120 static inline void oe_set_delta_child(struct packing_data
*pack
,
121 struct object_entry
*e
,
122 struct object_entry
*delta
)
125 e
->delta_child_idx
= (delta
- pack
->objects
) + 1;
127 e
->delta_child_idx
= 0;
130 static inline void oe_set_delta_sibling(struct packing_data
*pack
,
131 struct object_entry
*e
,
132 struct object_entry
*delta
)
135 e
->delta_sibling_idx
= (delta
- pack
->objects
) + 1;
137 e
->delta_sibling_idx
= 0;
140 static inline void oe_set_size(struct packing_data
*pack
,
141 struct object_entry
*e
,
144 if (size
< pack
->oe_size_limit
) {
149 if (oe_get_size_slow(pack
, e
) != size
)
150 BUG("'size' is supposed to be the object size!");
154 static inline void oe_set_delta_size(struct packing_data
*pack
,
155 struct object_entry
*e
,
158 if (size
< pack
->oe_delta_size_limit
) {
159 e
->delta_size_
= size
;
160 e
->delta_size_valid
= 1;
162 packing_data_lock(pack
);
163 if (!pack
->delta_size
)
164 ALLOC_ARRAY(pack
->delta_size
, pack
->nr_alloc
);
165 packing_data_unlock(pack
);
167 pack
->delta_size
[e
- pack
->objects
] = size
;
168 e
->delta_size_valid
= 0;
172 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
173 #define SIZE(obj) oe_size(&to_pack, obj)
174 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
175 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
176 #define DELTA(obj) oe_delta(&to_pack, obj)
177 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
178 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
179 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
180 #define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
181 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
182 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
183 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
185 static const char *pack_usage
[] = {
186 N_("git pack-objects --stdout [<options>] [< <ref-list> | < <object-list>]"),
187 N_("git pack-objects [<options>] <base-name> [< <ref-list> | < <object-list>]"),
191 static struct pack_idx_entry
**written_list
;
192 static uint32_t nr_result
, nr_written
, nr_seen
;
193 static struct bitmap_index
*bitmap_git
;
194 static uint32_t write_layer
;
196 static int non_empty
;
197 static int reuse_delta
= 1, reuse_object
= 1;
198 static int keep_unreachable
, unpack_unreachable
, include_tag
;
199 static timestamp_t unpack_unreachable_expiration
;
200 static int pack_loose_unreachable
;
202 static timestamp_t cruft_expiration
;
204 static int have_non_local_packs
;
205 static int incremental
;
206 static int ignore_packed_keep_on_disk
;
207 static int ignore_packed_keep_in_core
;
208 static int allow_ofs_delta
;
209 static struct pack_idx_option pack_idx_opts
;
210 static const char *base_name
;
211 static int progress
= 1;
212 static int window
= 10;
213 static unsigned long pack_size_limit
;
214 static int depth
= 50;
215 static int delta_search_threads
;
216 static int pack_to_stdout
;
219 static int num_preferred_base
;
220 static struct progress
*progress_state
;
222 static struct packed_git
*reuse_packfile
;
223 static uint32_t reuse_packfile_objects
;
224 static struct bitmap
*reuse_packfile_bitmap
;
226 static int use_bitmap_index_default
= 1;
227 static int use_bitmap_index
= -1;
228 static int allow_pack_reuse
= 1;
230 WRITE_BITMAP_FALSE
= 0,
233 } write_bitmap_index
;
234 static uint16_t write_bitmap_options
= BITMAP_OPT_HASH_CACHE
;
236 static int exclude_promisor_objects
;
238 static int use_delta_islands
;
240 static unsigned long delta_cache_size
= 0;
241 static unsigned long max_delta_cache_size
= DEFAULT_DELTA_CACHE_SIZE
;
242 static unsigned long cache_max_small_delta_size
= 1000;
244 static unsigned long window_memory_limit
= 0;
246 static struct string_list uri_protocols
= STRING_LIST_INIT_NODUP
;
248 enum missing_action
{
249 MA_ERROR
= 0, /* fail if any missing objects are encountered */
250 MA_ALLOW_ANY
, /* silently allow ALL missing objects */
251 MA_ALLOW_PROMISOR
, /* silently allow all missing PROMISOR objects */
253 static enum missing_action arg_missing_action
;
254 static show_object_fn fn_show_object
;
256 struct configured_exclusion
{
257 struct oidmap_entry e
;
261 static struct oidmap configured_exclusions
;
263 static struct oidset excluded_by_config
;
268 static uint32_t written
, written_delta
;
269 static uint32_t reused
, reused_delta
;
274 static struct commit
**indexed_commits
;
275 static unsigned int indexed_commits_nr
;
276 static unsigned int indexed_commits_alloc
;
278 static void index_commit_for_bitmap(struct commit
*commit
)
280 if (indexed_commits_nr
>= indexed_commits_alloc
) {
281 indexed_commits_alloc
= (indexed_commits_alloc
+ 32) * 2;
282 REALLOC_ARRAY(indexed_commits
, indexed_commits_alloc
);
285 indexed_commits
[indexed_commits_nr
++] = commit
;
288 static void *get_delta(struct object_entry
*entry
)
290 unsigned long size
, base_size
, delta_size
;
291 void *buf
, *base_buf
, *delta_buf
;
292 enum object_type type
;
294 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
296 die(_("unable to read %s"), oid_to_hex(&entry
->idx
.oid
));
297 base_buf
= read_object_file(&DELTA(entry
)->idx
.oid
, &type
,
300 die("unable to read %s",
301 oid_to_hex(&DELTA(entry
)->idx
.oid
));
302 delta_buf
= diff_delta(base_buf
, base_size
,
303 buf
, size
, &delta_size
, 0);
305 * We successfully computed this delta once but dropped it for
306 * memory reasons. Something is very wrong if this time we
307 * recompute and create a different delta.
309 if (!delta_buf
|| delta_size
!= DELTA_SIZE(entry
))
310 BUG("delta size changed");
316 static unsigned long do_compress(void **pptr
, unsigned long size
)
320 unsigned long maxsize
;
322 git_deflate_init(&stream
, pack_compression_level
);
323 maxsize
= git_deflate_bound(&stream
, size
);
326 out
= xmalloc(maxsize
);
330 stream
.avail_in
= size
;
331 stream
.next_out
= out
;
332 stream
.avail_out
= maxsize
;
333 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
335 git_deflate_end(&stream
);
338 return stream
.total_out
;
341 static unsigned long write_large_blob_data(struct git_istream
*st
, struct hashfile
*f
,
342 const struct object_id
*oid
)
345 unsigned char ibuf
[1024 * 16];
346 unsigned char obuf
[1024 * 16];
347 unsigned long olen
= 0;
349 git_deflate_init(&stream
, pack_compression_level
);
354 readlen
= read_istream(st
, ibuf
, sizeof(ibuf
));
356 die(_("unable to read %s"), oid_to_hex(oid
));
358 stream
.next_in
= ibuf
;
359 stream
.avail_in
= readlen
;
360 while ((stream
.avail_in
|| readlen
== 0) &&
361 (zret
== Z_OK
|| zret
== Z_BUF_ERROR
)) {
362 stream
.next_out
= obuf
;
363 stream
.avail_out
= sizeof(obuf
);
364 zret
= git_deflate(&stream
, readlen
? 0 : Z_FINISH
);
365 hashwrite(f
, obuf
, stream
.next_out
- obuf
);
366 olen
+= stream
.next_out
- obuf
;
369 die(_("deflate error (%d)"), zret
);
371 if (zret
!= Z_STREAM_END
)
372 die(_("deflate error (%d)"), zret
);
376 git_deflate_end(&stream
);
381 * we are going to reuse the existing object data as is. make
382 * sure it is not corrupt.
384 static int check_pack_inflate(struct packed_git
*p
,
385 struct pack_window
**w_curs
,
388 unsigned long expect
)
391 unsigned char fakebuf
[4096], *in
;
394 memset(&stream
, 0, sizeof(stream
));
395 git_inflate_init(&stream
);
397 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
399 stream
.next_out
= fakebuf
;
400 stream
.avail_out
= sizeof(fakebuf
);
401 st
= git_inflate(&stream
, Z_FINISH
);
402 offset
+= stream
.next_in
- in
;
403 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
404 git_inflate_end(&stream
);
405 return (st
== Z_STREAM_END
&&
406 stream
.total_out
== expect
&&
407 stream
.total_in
== len
) ? 0 : -1;
410 static void copy_pack_data(struct hashfile
*f
,
411 struct packed_git
*p
,
412 struct pack_window
**w_curs
,
420 in
= use_pack(p
, w_curs
, offset
, &avail
);
422 avail
= (unsigned long)len
;
423 hashwrite(f
, in
, avail
);
429 static inline int oe_size_greater_than(struct packing_data
*pack
,
430 const struct object_entry
*lhs
,
434 return lhs
->size_
> rhs
;
435 if (rhs
< pack
->oe_size_limit
) /* rhs < 2^x <= lhs ? */
437 return oe_get_size_slow(pack
, lhs
) > rhs
;
440 /* Return 0 if we will bust the pack-size limit */
441 static unsigned long write_no_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
442 unsigned long limit
, int usable_delta
)
444 unsigned long size
, datalen
;
445 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
446 dheader
[MAX_PACK_OBJECT_HEADER
];
448 enum object_type type
;
450 struct git_istream
*st
= NULL
;
451 const unsigned hashsz
= the_hash_algo
->rawsz
;
454 if (oe_type(entry
) == OBJ_BLOB
&&
455 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
) &&
456 (st
= open_istream(the_repository
, &entry
->idx
.oid
, &type
,
457 &size
, NULL
)) != NULL
)
460 buf
= read_object_file(&entry
->idx
.oid
, &type
, &size
);
462 die(_("unable to read %s"),
463 oid_to_hex(&entry
->idx
.oid
));
466 * make sure no cached delta data remains from a
467 * previous attempt before a pack split occurred.
469 FREE_AND_NULL(entry
->delta_data
);
470 entry
->z_delta_size
= 0;
471 } else if (entry
->delta_data
) {
472 size
= DELTA_SIZE(entry
);
473 buf
= entry
->delta_data
;
474 entry
->delta_data
= NULL
;
475 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
476 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
478 buf
= get_delta(entry
);
479 size
= DELTA_SIZE(entry
);
480 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
481 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
484 if (st
) /* large blob case, just assume we don't compress well */
486 else if (entry
->z_delta_size
)
487 datalen
= entry
->z_delta_size
;
489 datalen
= do_compress(&buf
, size
);
492 * The object header is a byte of 'type' followed by zero or
493 * more bytes of length.
495 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
498 if (type
== OBJ_OFS_DELTA
) {
500 * Deltas with relative base contain an additional
501 * encoding of the relative offset for the delta
502 * base from this object's position in the pack.
504 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
505 unsigned pos
= sizeof(dheader
) - 1;
506 dheader
[pos
] = ofs
& 127;
508 dheader
[--pos
] = 128 | (--ofs
& 127);
509 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
515 hashwrite(f
, header
, hdrlen
);
516 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
517 hdrlen
+= sizeof(dheader
) - pos
;
518 } else if (type
== OBJ_REF_DELTA
) {
520 * Deltas with a base reference contain
521 * additional bytes for the base object ID.
523 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
529 hashwrite(f
, header
, hdrlen
);
530 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
533 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
539 hashwrite(f
, header
, hdrlen
);
542 datalen
= write_large_blob_data(st
, f
, &entry
->idx
.oid
);
545 hashwrite(f
, buf
, datalen
);
549 return hdrlen
+ datalen
;
552 /* Return 0 if we will bust the pack-size limit */
553 static off_t
write_reuse_object(struct hashfile
*f
, struct object_entry
*entry
,
554 unsigned long limit
, int usable_delta
)
556 struct packed_git
*p
= IN_PACK(entry
);
557 struct pack_window
*w_curs
= NULL
;
560 enum object_type type
= oe_type(entry
);
562 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
563 dheader
[MAX_PACK_OBJECT_HEADER
];
565 const unsigned hashsz
= the_hash_algo
->rawsz
;
566 unsigned long entry_size
= SIZE(entry
);
569 type
= (allow_ofs_delta
&& DELTA(entry
)->idx
.offset
) ?
570 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
571 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
574 offset
= entry
->in_pack_offset
;
575 if (offset_to_pack_pos(p
, offset
, &pos
) < 0)
576 die(_("write_reuse_object: could not locate %s, expected at "
577 "offset %"PRIuMAX
" in pack %s"),
578 oid_to_hex(&entry
->idx
.oid
), (uintmax_t)offset
,
580 datalen
= pack_pos_to_offset(p
, pos
+ 1) - offset
;
581 if (!pack_to_stdout
&& p
->index_version
> 1 &&
582 check_pack_crc(p
, &w_curs
, offset
, datalen
,
583 pack_pos_to_index(p
, pos
))) {
584 error(_("bad packed object CRC for %s"),
585 oid_to_hex(&entry
->idx
.oid
));
587 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
590 offset
+= entry
->in_pack_header_size
;
591 datalen
-= entry
->in_pack_header_size
;
593 if (!pack_to_stdout
&& p
->index_version
== 1 &&
594 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry_size
)) {
595 error(_("corrupt packed object for %s"),
596 oid_to_hex(&entry
->idx
.oid
));
598 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
601 if (type
== OBJ_OFS_DELTA
) {
602 off_t ofs
= entry
->idx
.offset
- DELTA(entry
)->idx
.offset
;
603 unsigned pos
= sizeof(dheader
) - 1;
604 dheader
[pos
] = ofs
& 127;
606 dheader
[--pos
] = 128 | (--ofs
& 127);
607 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ hashsz
>= limit
) {
611 hashwrite(f
, header
, hdrlen
);
612 hashwrite(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
613 hdrlen
+= sizeof(dheader
) - pos
;
615 } else if (type
== OBJ_REF_DELTA
) {
616 if (limit
&& hdrlen
+ hashsz
+ datalen
+ hashsz
>= limit
) {
620 hashwrite(f
, header
, hdrlen
);
621 hashwrite(f
, DELTA(entry
)->idx
.oid
.hash
, hashsz
);
625 if (limit
&& hdrlen
+ datalen
+ hashsz
>= limit
) {
629 hashwrite(f
, header
, hdrlen
);
631 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
634 return hdrlen
+ datalen
;
637 /* Return 0 if we will bust the pack-size limit */
638 static off_t
write_object(struct hashfile
*f
,
639 struct object_entry
*entry
,
644 int usable_delta
, to_reuse
;
649 /* apply size limit if limited packsize and not first object */
650 if (!pack_size_limit
|| !nr_written
)
652 else if (pack_size_limit
<= write_offset
)
654 * the earlier object did not fit the limit; avoid
655 * mistaking this with unlimited (i.e. limit = 0).
659 limit
= pack_size_limit
- write_offset
;
662 usable_delta
= 0; /* no delta */
663 else if (!pack_size_limit
)
664 usable_delta
= 1; /* unlimited packfile */
665 else if (DELTA(entry
)->idx
.offset
== (off_t
)-1)
666 usable_delta
= 0; /* base was written to another pack */
667 else if (DELTA(entry
)->idx
.offset
)
668 usable_delta
= 1; /* base already exists in this pack */
670 usable_delta
= 0; /* base could end up in another pack */
673 to_reuse
= 0; /* explicit */
674 else if (!IN_PACK(entry
))
675 to_reuse
= 0; /* can't reuse what we don't have */
676 else if (oe_type(entry
) == OBJ_REF_DELTA
||
677 oe_type(entry
) == OBJ_OFS_DELTA
)
678 /* check_object() decided it for us ... */
679 to_reuse
= usable_delta
;
680 /* ... but pack split may override that */
681 else if (oe_type(entry
) != entry
->in_pack_type
)
682 to_reuse
= 0; /* pack has delta which is unusable */
683 else if (DELTA(entry
))
684 to_reuse
= 0; /* we want to pack afresh */
686 to_reuse
= 1; /* we have it in-pack undeltified,
687 * and we do not need to deltify it.
691 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
693 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
701 entry
->idx
.crc32
= crc32_end(f
);
705 enum write_one_status
{
706 WRITE_ONE_SKIP
= -1, /* already written */
707 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
708 WRITE_ONE_WRITTEN
= 1, /* normal */
709 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
712 static enum write_one_status
write_one(struct hashfile
*f
,
713 struct object_entry
*e
,
720 * we set offset to 1 (which is an impossible value) to mark
721 * the fact that this object is involved in "write its base
722 * first before writing a deltified object" recursion.
724 recursing
= (e
->idx
.offset
== 1);
726 warning(_("recursive delta detected for object %s"),
727 oid_to_hex(&e
->idx
.oid
));
728 return WRITE_ONE_RECURSIVE
;
729 } else if (e
->idx
.offset
|| e
->preferred_base
) {
730 /* offset is non zero if object is written already. */
731 return WRITE_ONE_SKIP
;
734 /* if we are deltified, write out base object first. */
736 e
->idx
.offset
= 1; /* now recurse */
737 switch (write_one(f
, DELTA(e
), offset
)) {
738 case WRITE_ONE_RECURSIVE
:
739 /* we cannot depend on this one */
744 case WRITE_ONE_BREAK
:
745 e
->idx
.offset
= recursing
;
746 return WRITE_ONE_BREAK
;
750 e
->idx
.offset
= *offset
;
751 size
= write_object(f
, e
, *offset
);
753 e
->idx
.offset
= recursing
;
754 return WRITE_ONE_BREAK
;
756 written_list
[nr_written
++] = &e
->idx
;
758 /* make sure off_t is sufficiently large not to wrap */
759 if (signed_add_overflows(*offset
, size
))
760 die(_("pack too large for current definition of off_t"));
762 return WRITE_ONE_WRITTEN
;
765 static int mark_tagged(const char *path UNUSED
, const struct object_id
*oid
,
766 int flag UNUSED
, void *cb_data UNUSED
)
768 struct object_id peeled
;
769 struct object_entry
*entry
= packlist_find(&to_pack
, oid
);
773 if (!peel_iterated_oid(oid
, &peeled
)) {
774 entry
= packlist_find(&to_pack
, &peeled
);
781 static inline unsigned char oe_layer(struct packing_data
*pack
,
782 struct object_entry
*e
)
786 return pack
->layer
[e
- pack
->objects
];
789 static inline void add_to_write_order(struct object_entry
**wo
,
791 struct object_entry
*e
)
793 if (e
->filled
|| oe_layer(&to_pack
, e
) != write_layer
)
799 static void add_descendants_to_write_order(struct object_entry
**wo
,
801 struct object_entry
*e
)
803 int add_to_order
= 1;
806 struct object_entry
*s
;
807 /* add this node... */
808 add_to_write_order(wo
, endp
, e
);
809 /* all its siblings... */
810 for (s
= DELTA_SIBLING(e
); s
; s
= DELTA_SIBLING(s
)) {
811 add_to_write_order(wo
, endp
, s
);
814 /* drop down a level to add left subtree nodes if possible */
815 if (DELTA_CHILD(e
)) {
820 /* our sibling might have some children, it is next */
821 if (DELTA_SIBLING(e
)) {
822 e
= DELTA_SIBLING(e
);
825 /* go back to our parent node */
827 while (e
&& !DELTA_SIBLING(e
)) {
828 /* we're on the right side of a subtree, keep
829 * going up until we can go right again */
833 /* done- we hit our original root node */
836 /* pass it off to sibling at this level */
837 e
= DELTA_SIBLING(e
);
842 static void add_family_to_write_order(struct object_entry
**wo
,
844 struct object_entry
*e
)
846 struct object_entry
*root
;
848 for (root
= e
; DELTA(root
); root
= DELTA(root
))
850 add_descendants_to_write_order(wo
, endp
, root
);
853 static void compute_layer_order(struct object_entry
**wo
, unsigned int *wo_end
)
855 unsigned int i
, last_untagged
;
856 struct object_entry
*objects
= to_pack
.objects
;
858 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
859 if (objects
[i
].tagged
)
861 add_to_write_order(wo
, wo_end
, &objects
[i
]);
866 * Then fill all the tagged tips.
868 for (; i
< to_pack
.nr_objects
; i
++) {
869 if (objects
[i
].tagged
)
870 add_to_write_order(wo
, wo_end
, &objects
[i
]);
874 * And then all remaining commits and tags.
876 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
877 if (oe_type(&objects
[i
]) != OBJ_COMMIT
&&
878 oe_type(&objects
[i
]) != OBJ_TAG
)
880 add_to_write_order(wo
, wo_end
, &objects
[i
]);
884 * And then all the trees.
886 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
887 if (oe_type(&objects
[i
]) != OBJ_TREE
)
889 add_to_write_order(wo
, wo_end
, &objects
[i
]);
893 * Finally all the rest in really tight order
895 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
896 if (!objects
[i
].filled
&& oe_layer(&to_pack
, &objects
[i
]) == write_layer
)
897 add_family_to_write_order(wo
, wo_end
, &objects
[i
]);
901 static struct object_entry
**compute_write_order(void)
903 uint32_t max_layers
= 1;
904 unsigned int i
, wo_end
;
906 struct object_entry
**wo
;
907 struct object_entry
*objects
= to_pack
.objects
;
909 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
910 objects
[i
].tagged
= 0;
911 objects
[i
].filled
= 0;
912 SET_DELTA_CHILD(&objects
[i
], NULL
);
913 SET_DELTA_SIBLING(&objects
[i
], NULL
);
917 * Fully connect delta_child/delta_sibling network.
918 * Make sure delta_sibling is sorted in the original
921 for (i
= to_pack
.nr_objects
; i
> 0;) {
922 struct object_entry
*e
= &objects
[--i
];
925 /* Mark me as the first child */
926 e
->delta_sibling_idx
= DELTA(e
)->delta_child_idx
;
927 SET_DELTA_CHILD(DELTA(e
), e
);
931 * Mark objects that are at the tip of tags.
933 for_each_tag_ref(mark_tagged
, NULL
);
935 if (use_delta_islands
) {
936 max_layers
= compute_pack_layers(&to_pack
);
940 ALLOC_ARRAY(wo
, to_pack
.nr_objects
);
943 for (; write_layer
< max_layers
; ++write_layer
)
944 compute_layer_order(wo
, &wo_end
);
946 if (wo_end
!= to_pack
.nr_objects
)
947 die(_("ordered %u objects, expected %"PRIu32
),
948 wo_end
, to_pack
.nr_objects
);
955 * A reused set of objects. All objects in a chunk have the same
956 * relative position in the original packfile and the generated
960 static struct reused_chunk
{
961 /* The offset of the first object of this chunk in the original
964 /* The difference for "original" minus the offset of the first object of
965 * this chunk in the generated packfile. */
968 static int reused_chunks_nr
;
969 static int reused_chunks_alloc
;
971 static void record_reused_object(off_t where
, off_t offset
)
973 if (reused_chunks_nr
&& reused_chunks
[reused_chunks_nr
-1].difference
== offset
)
976 ALLOC_GROW(reused_chunks
, reused_chunks_nr
+ 1,
977 reused_chunks_alloc
);
978 reused_chunks
[reused_chunks_nr
].original
= where
;
979 reused_chunks
[reused_chunks_nr
].difference
= offset
;
984 * Binary search to find the chunk that "where" is in. Note
985 * that we're not looking for an exact match, just the first
986 * chunk that contains it (which implicitly ends at the start
989 static off_t
find_reused_offset(off_t where
)
991 int lo
= 0, hi
= reused_chunks_nr
;
993 int mi
= lo
+ ((hi
- lo
) / 2);
994 if (where
== reused_chunks
[mi
].original
)
995 return reused_chunks
[mi
].difference
;
996 if (where
< reused_chunks
[mi
].original
)
1003 * The first chunk starts at zero, so we can't have gone below
1007 return reused_chunks
[lo
-1].difference
;
1010 static void write_reused_pack_one(size_t pos
, struct hashfile
*out
,
1011 struct pack_window
**w_curs
)
1013 off_t offset
, next
, cur
;
1014 enum object_type type
;
1017 offset
= pack_pos_to_offset(reuse_packfile
, pos
);
1018 next
= pack_pos_to_offset(reuse_packfile
, pos
+ 1);
1020 record_reused_object(offset
, offset
- hashfile_total(out
));
1023 type
= unpack_object_header(reuse_packfile
, w_curs
, &cur
, &size
);
1026 if (type
== OBJ_OFS_DELTA
) {
1030 unsigned char header
[MAX_PACK_OBJECT_HEADER
];
1033 base_offset
= get_delta_base(reuse_packfile
, w_curs
, &cur
, type
, offset
);
1034 assert(base_offset
!= 0);
1036 /* Convert to REF_DELTA if we must... */
1037 if (!allow_ofs_delta
) {
1039 struct object_id base_oid
;
1041 if (offset_to_pack_pos(reuse_packfile
, base_offset
, &base_pos
) < 0)
1042 die(_("expected object at offset %"PRIuMAX
" "
1044 (uintmax_t)base_offset
,
1045 reuse_packfile
->pack_name
);
1047 nth_packed_object_id(&base_oid
, reuse_packfile
,
1048 pack_pos_to_index(reuse_packfile
, base_pos
));
1050 len
= encode_in_pack_object_header(header
, sizeof(header
),
1051 OBJ_REF_DELTA
, size
);
1052 hashwrite(out
, header
, len
);
1053 hashwrite(out
, base_oid
.hash
, the_hash_algo
->rawsz
);
1054 copy_pack_data(out
, reuse_packfile
, w_curs
, cur
, next
- cur
);
1058 /* Otherwise see if we need to rewrite the offset... */
1059 fixup
= find_reused_offset(offset
) -
1060 find_reused_offset(base_offset
);
1062 unsigned char ofs_header
[10];
1063 unsigned i
, ofs_len
;
1064 off_t ofs
= offset
- base_offset
- fixup
;
1066 len
= encode_in_pack_object_header(header
, sizeof(header
),
1067 OBJ_OFS_DELTA
, size
);
1069 i
= sizeof(ofs_header
) - 1;
1070 ofs_header
[i
] = ofs
& 127;
1072 ofs_header
[--i
] = 128 | (--ofs
& 127);
1074 ofs_len
= sizeof(ofs_header
) - i
;
1076 hashwrite(out
, header
, len
);
1077 hashwrite(out
, ofs_header
+ sizeof(ofs_header
) - ofs_len
, ofs_len
);
1078 copy_pack_data(out
, reuse_packfile
, w_curs
, cur
, next
- cur
);
1082 /* ...otherwise we have no fixup, and can write it verbatim */
1085 copy_pack_data(out
, reuse_packfile
, w_curs
, offset
, next
- offset
);
1088 static size_t write_reused_pack_verbatim(struct hashfile
*out
,
1089 struct pack_window
**w_curs
)
1093 while (pos
< reuse_packfile_bitmap
->word_alloc
&&
1094 reuse_packfile_bitmap
->words
[pos
] == (eword_t
)~0)
1100 written
= (pos
* BITS_IN_EWORD
);
1101 to_write
= pack_pos_to_offset(reuse_packfile
, written
)
1102 - sizeof(struct pack_header
);
1104 /* We're recording one chunk, not one object. */
1105 record_reused_object(sizeof(struct pack_header
), 0);
1107 copy_pack_data(out
, reuse_packfile
, w_curs
,
1108 sizeof(struct pack_header
), to_write
);
1110 display_progress(progress_state
, written
);
1115 static void write_reused_pack(struct hashfile
*f
)
1119 struct pack_window
*w_curs
= NULL
;
1121 if (allow_ofs_delta
)
1122 i
= write_reused_pack_verbatim(f
, &w_curs
);
1124 for (; i
< reuse_packfile_bitmap
->word_alloc
; ++i
) {
1125 eword_t word
= reuse_packfile_bitmap
->words
[i
];
1126 size_t pos
= (i
* BITS_IN_EWORD
);
1128 for (offset
= 0; offset
< BITS_IN_EWORD
; ++offset
) {
1129 if ((word
>> offset
) == 0)
1132 offset
+= ewah_bit_ctz64(word
>> offset
);
1134 * Can use bit positions directly, even for MIDX
1135 * bitmaps. See comment in try_partial_reuse()
1138 write_reused_pack_one(pos
+ offset
, f
, &w_curs
);
1139 display_progress(progress_state
, ++written
);
1143 unuse_pack(&w_curs
);
1146 static void write_excluded_by_configs(void)
1148 struct oidset_iter iter
;
1149 const struct object_id
*oid
;
1151 oidset_iter_init(&excluded_by_config
, &iter
);
1152 while ((oid
= oidset_iter_next(&iter
))) {
1153 struct configured_exclusion
*ex
=
1154 oidmap_get(&configured_exclusions
, oid
);
1157 BUG("configured exclusion wasn't configured");
1158 write_in_full(1, ex
->pack_hash_hex
, strlen(ex
->pack_hash_hex
));
1159 write_in_full(1, " ", 1);
1160 write_in_full(1, ex
->uri
, strlen(ex
->uri
));
1161 write_in_full(1, "\n", 1);
1165 static const char no_split_warning
[] = N_(
1166 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
1169 static void write_pack_file(void)
1174 uint32_t nr_remaining
= nr_result
;
1175 time_t last_mtime
= 0;
1176 struct object_entry
**write_order
;
1178 if (progress
> pack_to_stdout
)
1179 progress_state
= start_progress(_("Writing objects"), nr_result
);
1180 ALLOC_ARRAY(written_list
, to_pack
.nr_objects
);
1181 write_order
= compute_write_order();
1184 unsigned char hash
[GIT_MAX_RAWSZ
];
1185 char *pack_tmp_name
= NULL
;
1188 f
= hashfd_throughput(1, "<stdout>", progress_state
);
1190 f
= create_tmp_packfile(&pack_tmp_name
);
1192 offset
= write_pack_header(f
, nr_remaining
);
1194 if (reuse_packfile
) {
1195 assert(pack_to_stdout
);
1196 write_reused_pack(f
);
1197 offset
= hashfile_total(f
);
1201 for (; i
< to_pack
.nr_objects
; i
++) {
1202 struct object_entry
*e
= write_order
[i
];
1203 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
1205 display_progress(progress_state
, written
);
1208 if (pack_to_stdout
) {
1210 * We never fsync when writing to stdout since we may
1211 * not be writing to an actual pack file. For instance,
1212 * the upload-pack code passes a pipe here. Calling
1213 * fsync on a pipe results in unnecessary
1214 * synchronization with the reader on some platforms.
1216 finalize_hashfile(f
, hash
, FSYNC_COMPONENT_NONE
,
1217 CSUM_HASH_IN_STREAM
| CSUM_CLOSE
);
1218 } else if (nr_written
== nr_remaining
) {
1219 finalize_hashfile(f
, hash
, FSYNC_COMPONENT_PACK
,
1220 CSUM_HASH_IN_STREAM
| CSUM_FSYNC
| CSUM_CLOSE
);
1223 * If we wrote the wrong number of entries in the
1224 * header, rewrite it like in fast-import.
1227 int fd
= finalize_hashfile(f
, hash
, FSYNC_COMPONENT_PACK
, 0);
1228 fixup_pack_header_footer(fd
, hash
, pack_tmp_name
,
1229 nr_written
, hash
, offset
);
1231 if (write_bitmap_index
) {
1232 if (write_bitmap_index
!= WRITE_BITMAP_QUIET
)
1233 warning(_(no_split_warning
));
1234 write_bitmap_index
= 0;
1238 if (!pack_to_stdout
) {
1240 struct strbuf tmpname
= STRBUF_INIT
;
1241 char *idx_tmp_name
= NULL
;
1244 * Packs are runtime accessed in their mtime
1245 * order since newer packs are more likely to contain
1246 * younger objects. So if we are creating multiple
1247 * packs then we should modify the mtime of later ones
1248 * to preserve this property.
1250 if (stat(pack_tmp_name
, &st
) < 0) {
1251 warning_errno(_("failed to stat %s"), pack_tmp_name
);
1252 } else if (!last_mtime
) {
1253 last_mtime
= st
.st_mtime
;
1256 utb
.actime
= st
.st_atime
;
1257 utb
.modtime
= --last_mtime
;
1258 if (utime(pack_tmp_name
, &utb
) < 0)
1259 warning_errno(_("failed utime() on %s"), pack_tmp_name
);
1262 strbuf_addf(&tmpname
, "%s-%s.", base_name
,
1265 if (write_bitmap_index
) {
1266 bitmap_writer_set_checksum(hash
);
1267 bitmap_writer_build_type_index(
1268 &to_pack
, written_list
, nr_written
);
1272 pack_idx_opts
.flags
|= WRITE_MTIMES
;
1274 stage_tmp_packfiles(&tmpname
, pack_tmp_name
,
1275 written_list
, nr_written
,
1276 &to_pack
, &pack_idx_opts
, hash
,
1279 if (write_bitmap_index
) {
1280 size_t tmpname_len
= tmpname
.len
;
1282 strbuf_addstr(&tmpname
, "bitmap");
1283 stop_progress(&progress_state
);
1285 bitmap_writer_show_progress(progress
);
1286 bitmap_writer_select_commits(indexed_commits
, indexed_commits_nr
, -1);
1287 if (bitmap_writer_build(&to_pack
) < 0)
1288 die(_("failed to write bitmap index"));
1289 bitmap_writer_finish(written_list
, nr_written
,
1290 tmpname
.buf
, write_bitmap_options
);
1291 write_bitmap_index
= 0;
1292 strbuf_setlen(&tmpname
, tmpname_len
);
1295 rename_tmp_packfile_idx(&tmpname
, &idx_tmp_name
);
1298 strbuf_release(&tmpname
);
1299 free(pack_tmp_name
);
1300 puts(hash_to_hex(hash
));
1303 /* mark written objects as written to previous pack */
1304 for (j
= 0; j
< nr_written
; j
++) {
1305 written_list
[j
]->offset
= (off_t
)-1;
1307 nr_remaining
-= nr_written
;
1308 } while (nr_remaining
&& i
< to_pack
.nr_objects
);
1312 stop_progress(&progress_state
);
1313 if (written
!= nr_result
)
1314 die(_("wrote %"PRIu32
" objects while expecting %"PRIu32
),
1315 written
, nr_result
);
1316 trace2_data_intmax("pack-objects", the_repository
,
1317 "write_pack_file/wrote", nr_result
);
1320 static int no_try_delta(const char *path
)
1322 static struct attr_check
*check
;
1325 check
= attr_check_initl("delta", NULL
);
1326 git_check_attr(the_repository
->index
, NULL
, path
, check
);
1327 if (ATTR_FALSE(check
->items
[0].value
))
1333 * When adding an object, check whether we have already added it
1334 * to our packing list. If so, we can skip. However, if we are
1335 * being asked to excludei t, but the previous mention was to include
1336 * it, make sure to adjust its flags and tweak our numbers accordingly.
1338 * As an optimization, we pass out the index position where we would have
1339 * found the item, since that saves us from having to look it up again a
1340 * few lines later when we want to add the new entry.
1342 static int have_duplicate_entry(const struct object_id
*oid
,
1345 struct object_entry
*entry
;
1347 if (reuse_packfile_bitmap
&&
1348 bitmap_walk_contains(bitmap_git
, reuse_packfile_bitmap
, oid
))
1351 entry
= packlist_find(&to_pack
, oid
);
1356 if (!entry
->preferred_base
)
1358 entry
->preferred_base
= 1;
1364 static int want_found_object(const struct object_id
*oid
, int exclude
,
1365 struct packed_git
*p
)
1372 if (!is_pack_valid(p
))
1376 * When asked to do --local (do not include an object that appears in a
1377 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1378 * an object that appears in a pack marked with .keep), finding a pack
1379 * that matches the criteria is sufficient for us to decide to omit it.
1380 * However, even if this pack does not satisfy the criteria, we need to
1381 * make sure no copy of this object appears in _any_ pack that makes us
1382 * to omit the object, so we need to check all the packs.
1384 * We can however first check whether these options can possibly matter;
1385 * if they do not matter we know we want the object in generated pack.
1386 * Otherwise, we signal "-1" at the end to tell the caller that we do
1387 * not know either way, and it needs to check more packs.
1391 * Objects in packs borrowed from elsewhere are discarded regardless of
1392 * if they appear in other packs that weren't borrowed.
1394 if (local
&& !p
->pack_local
)
1398 * Then handle .keep first, as we have a fast(er) path there.
1400 if (ignore_packed_keep_on_disk
|| ignore_packed_keep_in_core
) {
1402 * Set the flags for the kept-pack cache to be the ones we want
1405 * That is, if we are ignoring objects in on-disk keep packs,
1406 * then we want to search through the on-disk keep and ignore
1410 if (ignore_packed_keep_on_disk
)
1411 flags
|= ON_DISK_KEEP_PACKS
;
1412 if (ignore_packed_keep_in_core
)
1413 flags
|= IN_CORE_KEEP_PACKS
;
1415 if (ignore_packed_keep_on_disk
&& p
->pack_keep
)
1417 if (ignore_packed_keep_in_core
&& p
->pack_keep_in_core
)
1419 if (has_object_kept_pack(oid
, flags
))
1424 * At this point we know definitively that either we don't care about
1425 * keep-packs, or the object is not in one. Keep checking other
1428 if (!local
|| !have_non_local_packs
)
1431 /* we don't know yet; keep looking for more packs */
1435 static int want_object_in_pack_one(struct packed_git
*p
,
1436 const struct object_id
*oid
,
1438 struct packed_git
**found_pack
,
1439 off_t
*found_offset
)
1443 if (p
== *found_pack
)
1444 offset
= *found_offset
;
1446 offset
= find_pack_entry_one(oid
->hash
, p
);
1450 if (!is_pack_valid(p
))
1452 *found_offset
= offset
;
1455 return want_found_object(oid
, exclude
, p
);
1461 * Check whether we want the object in the pack (e.g., we do not want
1462 * objects found in non-local stores if the "--local" option was used).
1464 * If the caller already knows an existing pack it wants to take the object
1465 * from, that is passed in *found_pack and *found_offset; otherwise this
1466 * function finds if there is any pack that has the object and returns the pack
1467 * and its offset in these variables.
1469 static int want_object_in_pack(const struct object_id
*oid
,
1471 struct packed_git
**found_pack
,
1472 off_t
*found_offset
)
1475 struct list_head
*pos
;
1476 struct multi_pack_index
*m
;
1478 if (!exclude
&& local
&& has_loose_object_nonlocal(oid
))
1482 * If we already know the pack object lives in, start checks from that
1483 * pack - in the usual case when neither --local was given nor .keep files
1484 * are present we will determine the answer right now.
1487 want
= want_found_object(oid
, exclude
, *found_pack
);
1495 for (m
= get_multi_pack_index(the_repository
); m
; m
= m
->next
) {
1496 struct pack_entry e
;
1497 if (fill_midx_entry(the_repository
, oid
, &e
, m
)) {
1498 want
= want_object_in_pack_one(e
.p
, oid
, exclude
, found_pack
, found_offset
);
1504 list_for_each(pos
, get_packed_git_mru(the_repository
)) {
1505 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1506 want
= want_object_in_pack_one(p
, oid
, exclude
, found_pack
, found_offset
);
1507 if (!exclude
&& want
> 0)
1509 get_packed_git_mru(the_repository
));
1514 if (uri_protocols
.nr
) {
1515 struct configured_exclusion
*ex
=
1516 oidmap_get(&configured_exclusions
, oid
);
1521 for (i
= 0; i
< uri_protocols
.nr
; i
++) {
1522 if (skip_prefix(ex
->uri
,
1523 uri_protocols
.items
[i
].string
,
1526 oidset_insert(&excluded_by_config
, oid
);
1536 static struct object_entry
*create_object_entry(const struct object_id
*oid
,
1537 enum object_type type
,
1541 struct packed_git
*found_pack
,
1544 struct object_entry
*entry
;
1546 entry
= packlist_alloc(&to_pack
, oid
);
1548 oe_set_type(entry
, type
);
1550 entry
->preferred_base
= 1;
1554 oe_set_in_pack(&to_pack
, entry
, found_pack
);
1555 entry
->in_pack_offset
= found_offset
;
1558 entry
->no_try_delta
= no_try_delta
;
1563 static const char no_closure_warning
[] = N_(
1564 "disabling bitmap writing, as some objects are not being packed"
1567 static int add_object_entry(const struct object_id
*oid
, enum object_type type
,
1568 const char *name
, int exclude
)
1570 struct packed_git
*found_pack
= NULL
;
1571 off_t found_offset
= 0;
1573 display_progress(progress_state
, ++nr_seen
);
1575 if (have_duplicate_entry(oid
, exclude
))
1578 if (!want_object_in_pack(oid
, exclude
, &found_pack
, &found_offset
)) {
1579 /* The pack is missing an object, so it will not have closure */
1580 if (write_bitmap_index
) {
1581 if (write_bitmap_index
!= WRITE_BITMAP_QUIET
)
1582 warning(_(no_closure_warning
));
1583 write_bitmap_index
= 0;
1588 create_object_entry(oid
, type
, pack_name_hash(name
),
1589 exclude
, name
&& no_try_delta(name
),
1590 found_pack
, found_offset
);
1594 static int add_object_entry_from_bitmap(const struct object_id
*oid
,
1595 enum object_type type
,
1596 int flags UNUSED
, uint32_t name_hash
,
1597 struct packed_git
*pack
, off_t offset
)
1599 display_progress(progress_state
, ++nr_seen
);
1601 if (have_duplicate_entry(oid
, 0))
1604 if (!want_object_in_pack(oid
, 0, &pack
, &offset
))
1607 create_object_entry(oid
, type
, name_hash
, 0, 0, pack
, offset
);
1611 struct pbase_tree_cache
{
1612 struct object_id oid
;
1616 unsigned long tree_size
;
1619 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1620 static int pbase_tree_cache_ix(const struct object_id
*oid
)
1622 return oid
->hash
[0] % ARRAY_SIZE(pbase_tree_cache
);
1624 static int pbase_tree_cache_ix_incr(int ix
)
1626 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1629 static struct pbase_tree
{
1630 struct pbase_tree
*next
;
1631 /* This is a phony "cache" entry; we are not
1632 * going to evict it or find it through _get()
1633 * mechanism -- this is for the toplevel node that
1634 * would almost always change with any commit.
1636 struct pbase_tree_cache pcache
;
1639 static struct pbase_tree_cache
*pbase_tree_get(const struct object_id
*oid
)
1641 struct pbase_tree_cache
*ent
, *nent
;
1644 enum object_type type
;
1646 int my_ix
= pbase_tree_cache_ix(oid
);
1647 int available_ix
= -1;
1649 /* pbase-tree-cache acts as a limited hashtable.
1650 * your object will be found at your index or within a few
1651 * slots after that slot if it is cached.
1653 for (neigh
= 0; neigh
< 8; neigh
++) {
1654 ent
= pbase_tree_cache
[my_ix
];
1655 if (ent
&& oideq(&ent
->oid
, oid
)) {
1659 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1660 ((0 <= available_ix
) &&
1661 (!ent
&& pbase_tree_cache
[available_ix
])))
1662 available_ix
= my_ix
;
1665 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1668 /* Did not find one. Either we got a bogus request or
1669 * we need to read and perhaps cache.
1671 data
= read_object_file(oid
, &type
, &size
);
1674 if (type
!= OBJ_TREE
) {
1679 /* We need to either cache or return a throwaway copy */
1681 if (available_ix
< 0)
1684 ent
= pbase_tree_cache
[available_ix
];
1685 my_ix
= available_ix
;
1689 nent
= xmalloc(sizeof(*nent
));
1690 nent
->temporary
= (available_ix
< 0);
1693 /* evict and reuse */
1694 free(ent
->tree_data
);
1697 oidcpy(&nent
->oid
, oid
);
1698 nent
->tree_data
= data
;
1699 nent
->tree_size
= size
;
1701 if (!nent
->temporary
)
1702 pbase_tree_cache
[my_ix
] = nent
;
1706 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1708 if (!cache
->temporary
) {
1712 free(cache
->tree_data
);
1716 static size_t name_cmp_len(const char *name
)
1718 return strcspn(name
, "\n/");
1721 static void add_pbase_object(struct tree_desc
*tree
,
1724 const char *fullname
)
1726 struct name_entry entry
;
1729 while (tree_entry(tree
,&entry
)) {
1730 if (S_ISGITLINK(entry
.mode
))
1732 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1733 memcmp(name
, entry
.path
, cmplen
);
1738 if (name
[cmplen
] != '/') {
1739 add_object_entry(&entry
.oid
,
1740 object_type(entry
.mode
),
1744 if (S_ISDIR(entry
.mode
)) {
1745 struct tree_desc sub
;
1746 struct pbase_tree_cache
*tree
;
1747 const char *down
= name
+cmplen
+1;
1748 size_t downlen
= name_cmp_len(down
);
1750 tree
= pbase_tree_get(&entry
.oid
);
1753 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1755 add_pbase_object(&sub
, down
, downlen
, fullname
);
1756 pbase_tree_put(tree
);
1761 static unsigned *done_pbase_paths
;
1762 static int done_pbase_paths_num
;
1763 static int done_pbase_paths_alloc
;
1764 static int done_pbase_path_pos(unsigned hash
)
1767 int hi
= done_pbase_paths_num
;
1769 int mi
= lo
+ (hi
- lo
) / 2;
1770 if (done_pbase_paths
[mi
] == hash
)
1772 if (done_pbase_paths
[mi
] < hash
)
1780 static int check_pbase_path(unsigned hash
)
1782 int pos
= done_pbase_path_pos(hash
);
1786 ALLOC_GROW(done_pbase_paths
,
1787 done_pbase_paths_num
+ 1,
1788 done_pbase_paths_alloc
);
1789 done_pbase_paths_num
++;
1790 if (pos
< done_pbase_paths_num
)
1791 MOVE_ARRAY(done_pbase_paths
+ pos
+ 1, done_pbase_paths
+ pos
,
1792 done_pbase_paths_num
- pos
- 1);
1793 done_pbase_paths
[pos
] = hash
;
1797 static void add_preferred_base_object(const char *name
)
1799 struct pbase_tree
*it
;
1801 unsigned hash
= pack_name_hash(name
);
1803 if (!num_preferred_base
|| check_pbase_path(hash
))
1806 cmplen
= name_cmp_len(name
);
1807 for (it
= pbase_tree
; it
; it
= it
->next
) {
1809 add_object_entry(&it
->pcache
.oid
, OBJ_TREE
, NULL
, 1);
1812 struct tree_desc tree
;
1813 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1814 add_pbase_object(&tree
, name
, cmplen
, name
);
1819 static void add_preferred_base(struct object_id
*oid
)
1821 struct pbase_tree
*it
;
1824 struct object_id tree_oid
;
1826 if (window
<= num_preferred_base
++)
1829 data
= read_object_with_reference(the_repository
, oid
,
1830 OBJ_TREE
, &size
, &tree_oid
);
1834 for (it
= pbase_tree
; it
; it
= it
->next
) {
1835 if (oideq(&it
->pcache
.oid
, &tree_oid
)) {
1841 CALLOC_ARRAY(it
, 1);
1842 it
->next
= pbase_tree
;
1845 oidcpy(&it
->pcache
.oid
, &tree_oid
);
1846 it
->pcache
.tree_data
= data
;
1847 it
->pcache
.tree_size
= size
;
1850 static void cleanup_preferred_base(void)
1852 struct pbase_tree
*it
;
1858 struct pbase_tree
*tmp
= it
;
1860 free(tmp
->pcache
.tree_data
);
1864 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1865 if (!pbase_tree_cache
[i
])
1867 free(pbase_tree_cache
[i
]->tree_data
);
1868 FREE_AND_NULL(pbase_tree_cache
[i
]);
1871 FREE_AND_NULL(done_pbase_paths
);
1872 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1876 * Return 1 iff the object specified by "delta" can be sent
1877 * literally as a delta against the base in "base_sha1". If
1878 * so, then *base_out will point to the entry in our packing
1879 * list, or NULL if we must use the external-base list.
1881 * Depth value does not matter - find_deltas() will
1882 * never consider reused delta as the base object to
1883 * deltify other objects against, in order to avoid
1886 static int can_reuse_delta(const struct object_id
*base_oid
,
1887 struct object_entry
*delta
,
1888 struct object_entry
**base_out
)
1890 struct object_entry
*base
;
1893 * First see if we're already sending the base (or it's explicitly in
1894 * our "excluded" list).
1896 base
= packlist_find(&to_pack
, base_oid
);
1898 if (!in_same_island(&delta
->idx
.oid
, &base
->idx
.oid
))
1905 * Otherwise, reachability bitmaps may tell us if the receiver has it,
1906 * even if it was buried too deep in history to make it into the
1909 if (thin
&& bitmap_has_oid_in_uninteresting(bitmap_git
, base_oid
)) {
1910 if (use_delta_islands
) {
1911 if (!in_same_island(&delta
->idx
.oid
, base_oid
))
1921 static void prefetch_to_pack(uint32_t object_index_start
) {
1922 struct oid_array to_fetch
= OID_ARRAY_INIT
;
1925 for (i
= object_index_start
; i
< to_pack
.nr_objects
; i
++) {
1926 struct object_entry
*entry
= to_pack
.objects
+ i
;
1928 if (!oid_object_info_extended(the_repository
,
1931 OBJECT_INFO_FOR_PREFETCH
))
1933 oid_array_append(&to_fetch
, &entry
->idx
.oid
);
1935 promisor_remote_get_direct(the_repository
,
1936 to_fetch
.oid
, to_fetch
.nr
);
1937 oid_array_clear(&to_fetch
);
1940 static void check_object(struct object_entry
*entry
, uint32_t object_index
)
1942 unsigned long canonical_size
;
1943 enum object_type type
;
1944 struct object_info oi
= {.typep
= &type
, .sizep
= &canonical_size
};
1946 if (IN_PACK(entry
)) {
1947 struct packed_git
*p
= IN_PACK(entry
);
1948 struct pack_window
*w_curs
= NULL
;
1950 struct object_id base_ref
;
1951 struct object_entry
*base_entry
;
1952 unsigned long used
, used_0
;
1953 unsigned long avail
;
1955 unsigned char *buf
, c
;
1956 enum object_type type
;
1957 unsigned long in_pack_size
;
1959 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1962 * We want in_pack_type even if we do not reuse delta
1963 * since non-delta representations could still be reused.
1965 used
= unpack_object_header_buffer(buf
, avail
,
1972 BUG("invalid type %d", type
);
1973 entry
->in_pack_type
= type
;
1976 * Determine if this is a delta and if so whether we can
1977 * reuse it or not. Otherwise let's find out as cheaply as
1978 * possible what the actual type and size for this object is.
1980 switch (entry
->in_pack_type
) {
1982 /* Not a delta hence we've already got all we need. */
1983 oe_set_type(entry
, entry
->in_pack_type
);
1984 SET_SIZE(entry
, in_pack_size
);
1985 entry
->in_pack_header_size
= used
;
1986 if (oe_type(entry
) < OBJ_COMMIT
|| oe_type(entry
) > OBJ_BLOB
)
1988 unuse_pack(&w_curs
);
1991 if (reuse_delta
&& !entry
->preferred_base
) {
1993 use_pack(p
, &w_curs
,
1994 entry
->in_pack_offset
+ used
,
1998 entry
->in_pack_header_size
= used
+ the_hash_algo
->rawsz
;
2001 buf
= use_pack(p
, &w_curs
,
2002 entry
->in_pack_offset
+ used
, NULL
);
2008 if (!ofs
|| MSB(ofs
, 7)) {
2009 error(_("delta base offset overflow in pack for %s"),
2010 oid_to_hex(&entry
->idx
.oid
));
2014 ofs
= (ofs
<< 7) + (c
& 127);
2016 ofs
= entry
->in_pack_offset
- ofs
;
2017 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
2018 error(_("delta base offset out of bound for %s"),
2019 oid_to_hex(&entry
->idx
.oid
));
2022 if (reuse_delta
&& !entry
->preferred_base
) {
2024 if (offset_to_pack_pos(p
, ofs
, &pos
) < 0)
2026 if (!nth_packed_object_id(&base_ref
, p
,
2027 pack_pos_to_index(p
, pos
)))
2030 entry
->in_pack_header_size
= used
+ used_0
;
2035 can_reuse_delta(&base_ref
, entry
, &base_entry
)) {
2036 oe_set_type(entry
, entry
->in_pack_type
);
2037 SET_SIZE(entry
, in_pack_size
); /* delta size */
2038 SET_DELTA_SIZE(entry
, in_pack_size
);
2041 SET_DELTA(entry
, base_entry
);
2042 entry
->delta_sibling_idx
= base_entry
->delta_child_idx
;
2043 SET_DELTA_CHILD(base_entry
, entry
);
2045 SET_DELTA_EXT(entry
, &base_ref
);
2048 unuse_pack(&w_curs
);
2052 if (oe_type(entry
)) {
2056 * This must be a delta and we already know what the
2057 * final object type is. Let's extract the actual
2058 * object size from the delta header.
2060 delta_pos
= entry
->in_pack_offset
+ entry
->in_pack_header_size
;
2061 canonical_size
= get_size_from_delta(p
, &w_curs
, delta_pos
);
2062 if (canonical_size
== 0)
2064 SET_SIZE(entry
, canonical_size
);
2065 unuse_pack(&w_curs
);
2070 * No choice but to fall back to the recursive delta walk
2071 * with oid_object_info() to find about the object type
2075 unuse_pack(&w_curs
);
2078 if (oid_object_info_extended(the_repository
, &entry
->idx
.oid
, &oi
,
2079 OBJECT_INFO_SKIP_FETCH_OBJECT
| OBJECT_INFO_LOOKUP_REPLACE
) < 0) {
2080 if (has_promisor_remote()) {
2081 prefetch_to_pack(object_index
);
2082 if (oid_object_info_extended(the_repository
, &entry
->idx
.oid
, &oi
,
2083 OBJECT_INFO_SKIP_FETCH_OBJECT
| OBJECT_INFO_LOOKUP_REPLACE
) < 0)
2089 oe_set_type(entry
, type
);
2090 if (entry
->type_valid
) {
2091 SET_SIZE(entry
, canonical_size
);
2094 * Bad object type is checked in prepare_pack(). This is
2095 * to permit a missing preferred base object to be ignored
2096 * as a preferred base. Doing so can result in a larger
2097 * pack file, but the transfer will still take place.
2102 static int pack_offset_sort(const void *_a
, const void *_b
)
2104 const struct object_entry
*a
= *(struct object_entry
**)_a
;
2105 const struct object_entry
*b
= *(struct object_entry
**)_b
;
2106 const struct packed_git
*a_in_pack
= IN_PACK(a
);
2107 const struct packed_git
*b_in_pack
= IN_PACK(b
);
2109 /* avoid filesystem trashing with loose objects */
2110 if (!a_in_pack
&& !b_in_pack
)
2111 return oidcmp(&a
->idx
.oid
, &b
->idx
.oid
);
2113 if (a_in_pack
< b_in_pack
)
2115 if (a_in_pack
> b_in_pack
)
2117 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
2118 (a
->in_pack_offset
> b
->in_pack_offset
);
2122 * Drop an on-disk delta we were planning to reuse. Naively, this would
2123 * just involve blanking out the "delta" field, but we have to deal
2124 * with some extra book-keeping:
2126 * 1. Removing ourselves from the delta_sibling linked list.
2128 * 2. Updating our size/type to the non-delta representation. These were
2129 * either not recorded initially (size) or overwritten with the delta type
2130 * (type) when check_object() decided to reuse the delta.
2132 * 3. Resetting our delta depth, as we are now a base object.
2134 static void drop_reused_delta(struct object_entry
*entry
)
2136 unsigned *idx
= &to_pack
.objects
[entry
->delta_idx
- 1].delta_child_idx
;
2137 struct object_info oi
= OBJECT_INFO_INIT
;
2138 enum object_type type
;
2142 struct object_entry
*oe
= &to_pack
.objects
[*idx
- 1];
2145 *idx
= oe
->delta_sibling_idx
;
2147 idx
= &oe
->delta_sibling_idx
;
2149 SET_DELTA(entry
, NULL
);
2154 if (packed_object_info(the_repository
, IN_PACK(entry
), entry
->in_pack_offset
, &oi
) < 0) {
2156 * We failed to get the info from this pack for some reason;
2157 * fall back to oid_object_info, which may find another copy.
2158 * And if that fails, the error will be recorded in oe_type(entry)
2159 * and dealt with in prepare_pack().
2162 oid_object_info(the_repository
, &entry
->idx
.oid
, &size
));
2164 oe_set_type(entry
, type
);
2166 SET_SIZE(entry
, size
);
2170 * Follow the chain of deltas from this entry onward, throwing away any links
2171 * that cause us to hit a cycle (as determined by the DFS state flags in
2174 * We also detect too-long reused chains that would violate our --depth
2177 static void break_delta_chains(struct object_entry
*entry
)
2180 * The actual depth of each object we will write is stored as an int,
2181 * as it cannot exceed our int "depth" limit. But before we break
2182 * changes based no that limit, we may potentially go as deep as the
2183 * number of objects, which is elsewhere bounded to a uint32_t.
2185 uint32_t total_depth
;
2186 struct object_entry
*cur
, *next
;
2188 for (cur
= entry
, total_depth
= 0;
2190 cur
= DELTA(cur
), total_depth
++) {
2191 if (cur
->dfs_state
== DFS_DONE
) {
2193 * We've already seen this object and know it isn't
2194 * part of a cycle. We do need to append its depth
2197 total_depth
+= cur
->depth
;
2202 * We break cycles before looping, so an ACTIVE state (or any
2203 * other cruft which made its way into the state variable)
2206 if (cur
->dfs_state
!= DFS_NONE
)
2207 BUG("confusing delta dfs state in first pass: %d",
2211 * Now we know this is the first time we've seen the object. If
2212 * it's not a delta, we're done traversing, but we'll mark it
2213 * done to save time on future traversals.
2216 cur
->dfs_state
= DFS_DONE
;
2221 * Mark ourselves as active and see if the next step causes
2222 * us to cycle to another active object. It's important to do
2223 * this _before_ we loop, because it impacts where we make the
2224 * cut, and thus how our total_depth counter works.
2225 * E.g., We may see a partial loop like:
2227 * A -> B -> C -> D -> B
2229 * Cutting B->C breaks the cycle. But now the depth of A is
2230 * only 1, and our total_depth counter is at 3. The size of the
2231 * error is always one less than the size of the cycle we
2232 * broke. Commits C and D were "lost" from A's chain.
2234 * If we instead cut D->B, then the depth of A is correct at 3.
2235 * We keep all commits in the chain that we examined.
2237 cur
->dfs_state
= DFS_ACTIVE
;
2238 if (DELTA(cur
)->dfs_state
== DFS_ACTIVE
) {
2239 drop_reused_delta(cur
);
2240 cur
->dfs_state
= DFS_DONE
;
2246 * And now that we've gone all the way to the bottom of the chain, we
2247 * need to clear the active flags and set the depth fields as
2248 * appropriate. Unlike the loop above, which can quit when it drops a
2249 * delta, we need to keep going to look for more depth cuts. So we need
2250 * an extra "next" pointer to keep going after we reset cur->delta.
2252 for (cur
= entry
; cur
; cur
= next
) {
2256 * We should have a chain of zero or more ACTIVE states down to
2257 * a final DONE. We can quit after the DONE, because either it
2258 * has no bases, or we've already handled them in a previous
2261 if (cur
->dfs_state
== DFS_DONE
)
2263 else if (cur
->dfs_state
!= DFS_ACTIVE
)
2264 BUG("confusing delta dfs state in second pass: %d",
2268 * If the total_depth is more than depth, then we need to snip
2269 * the chain into two or more smaller chains that don't exceed
2270 * the maximum depth. Most of the resulting chains will contain
2271 * (depth + 1) entries (i.e., depth deltas plus one base), and
2272 * the last chain (i.e., the one containing entry) will contain
2273 * whatever entries are left over, namely
2274 * (total_depth % (depth + 1)) of them.
2276 * Since we are iterating towards decreasing depth, we need to
2277 * decrement total_depth as we go, and we need to write to the
2278 * entry what its final depth will be after all of the
2279 * snipping. Since we're snipping into chains of length (depth
2280 * + 1) entries, the final depth of an entry will be its
2281 * original depth modulo (depth + 1). Any time we encounter an
2282 * entry whose final depth is supposed to be zero, we snip it
2283 * from its delta base, thereby making it so.
2285 cur
->depth
= (total_depth
--) % (depth
+ 1);
2287 drop_reused_delta(cur
);
2289 cur
->dfs_state
= DFS_DONE
;
2293 static void get_object_details(void)
2296 struct object_entry
**sorted_by_offset
;
2299 progress_state
= start_progress(_("Counting objects"),
2300 to_pack
.nr_objects
);
2302 CALLOC_ARRAY(sorted_by_offset
, to_pack
.nr_objects
);
2303 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
2304 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
2305 QSORT(sorted_by_offset
, to_pack
.nr_objects
, pack_offset_sort
);
2307 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2308 struct object_entry
*entry
= sorted_by_offset
[i
];
2309 check_object(entry
, i
);
2310 if (entry
->type_valid
&&
2311 oe_size_greater_than(&to_pack
, entry
, big_file_threshold
))
2312 entry
->no_try_delta
= 1;
2313 display_progress(progress_state
, i
+ 1);
2315 stop_progress(&progress_state
);
2318 * This must happen in a second pass, since we rely on the delta
2319 * information for the whole list being completed.
2321 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
2322 break_delta_chains(&to_pack
.objects
[i
]);
2324 free(sorted_by_offset
);
2328 * We search for deltas in a list sorted by type, by filename hash, and then
2329 * by size, so that we see progressively smaller and smaller files.
2330 * That's because we prefer deltas to be from the bigger file
2331 * to the smaller -- deletes are potentially cheaper, but perhaps
2332 * more importantly, the bigger file is likely the more recent
2333 * one. The deepest deltas are therefore the oldest objects which are
2334 * less susceptible to be accessed often.
2336 static int type_size_sort(const void *_a
, const void *_b
)
2338 const struct object_entry
*a
= *(struct object_entry
**)_a
;
2339 const struct object_entry
*b
= *(struct object_entry
**)_b
;
2340 const enum object_type a_type
= oe_type(a
);
2341 const enum object_type b_type
= oe_type(b
);
2342 const unsigned long a_size
= SIZE(a
);
2343 const unsigned long b_size
= SIZE(b
);
2345 if (a_type
> b_type
)
2347 if (a_type
< b_type
)
2349 if (a
->hash
> b
->hash
)
2351 if (a
->hash
< b
->hash
)
2353 if (a
->preferred_base
> b
->preferred_base
)
2355 if (a
->preferred_base
< b
->preferred_base
)
2357 if (use_delta_islands
) {
2358 const int island_cmp
= island_delta_cmp(&a
->idx
.oid
, &b
->idx
.oid
);
2362 if (a_size
> b_size
)
2364 if (a_size
< b_size
)
2366 return a
< b
? -1 : (a
> b
); /* newest first */
2370 struct object_entry
*entry
;
2372 struct delta_index
*index
;
2376 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
2377 unsigned long delta_size
)
2379 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
2382 if (delta_size
< cache_max_small_delta_size
)
2385 /* cache delta, if objects are large enough compared to delta size */
2386 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
2392 /* Protect delta_cache_size */
2393 static pthread_mutex_t cache_mutex
;
2394 #define cache_lock() pthread_mutex_lock(&cache_mutex)
2395 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
2398 * Protect object list partitioning (e.g. struct thread_param) and
2401 static pthread_mutex_t progress_mutex
;
2402 #define progress_lock() pthread_mutex_lock(&progress_mutex)
2403 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
2406 * Access to struct object_entry is unprotected since each thread owns
2407 * a portion of the main object list. Just don't access object entries
2408 * ahead in the list because they can be stolen and would need
2409 * progress_mutex for protection.
2412 static inline int oe_size_less_than(struct packing_data
*pack
,
2413 const struct object_entry
*lhs
,
2416 if (lhs
->size_valid
)
2417 return lhs
->size_
< rhs
;
2418 if (rhs
< pack
->oe_size_limit
) /* rhs < 2^x <= lhs ? */
2420 return oe_get_size_slow(pack
, lhs
) < rhs
;
2423 static inline void oe_set_tree_depth(struct packing_data
*pack
,
2424 struct object_entry
*e
,
2425 unsigned int tree_depth
)
2427 if (!pack
->tree_depth
)
2428 CALLOC_ARRAY(pack
->tree_depth
, pack
->nr_alloc
);
2429 pack
->tree_depth
[e
- pack
->objects
] = tree_depth
;
2433 * Return the size of the object without doing any delta
2434 * reconstruction (so non-deltas are true object sizes, but deltas
2435 * return the size of the delta data).
2437 unsigned long oe_get_size_slow(struct packing_data
*pack
,
2438 const struct object_entry
*e
)
2440 struct packed_git
*p
;
2441 struct pack_window
*w_curs
;
2443 enum object_type type
;
2444 unsigned long used
, avail
, size
;
2446 if (e
->type_
!= OBJ_OFS_DELTA
&& e
->type_
!= OBJ_REF_DELTA
) {
2447 packing_data_lock(&to_pack
);
2448 if (oid_object_info(the_repository
, &e
->idx
.oid
, &size
) < 0)
2449 die(_("unable to get size of %s"),
2450 oid_to_hex(&e
->idx
.oid
));
2451 packing_data_unlock(&to_pack
);
2455 p
= oe_in_pack(pack
, e
);
2457 BUG("when e->type is a delta, it must belong to a pack");
2459 packing_data_lock(&to_pack
);
2461 buf
= use_pack(p
, &w_curs
, e
->in_pack_offset
, &avail
);
2462 used
= unpack_object_header_buffer(buf
, avail
, &type
, &size
);
2464 die(_("unable to parse object header of %s"),
2465 oid_to_hex(&e
->idx
.oid
));
2467 unuse_pack(&w_curs
);
2468 packing_data_unlock(&to_pack
);
2472 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
2473 unsigned max_depth
, unsigned long *mem_usage
)
2475 struct object_entry
*trg_entry
= trg
->entry
;
2476 struct object_entry
*src_entry
= src
->entry
;
2477 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
2479 enum object_type type
;
2482 /* Don't bother doing diffs between different types */
2483 if (oe_type(trg_entry
) != oe_type(src_entry
))
2487 * We do not bother to try a delta that we discarded on an
2488 * earlier try, but only when reusing delta data. Note that
2489 * src_entry that is marked as the preferred_base should always
2490 * be considered, as even if we produce a suboptimal delta against
2491 * it, we will still save the transfer cost, as we already know
2492 * the other side has it and we won't send src_entry at all.
2494 if (reuse_delta
&& IN_PACK(trg_entry
) &&
2495 IN_PACK(trg_entry
) == IN_PACK(src_entry
) &&
2496 !src_entry
->preferred_base
&&
2497 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
2498 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
2501 /* Let's not bust the allowed depth. */
2502 if (src
->depth
>= max_depth
)
2505 /* Now some size filtering heuristics. */
2506 trg_size
= SIZE(trg_entry
);
2507 if (!DELTA(trg_entry
)) {
2508 max_size
= trg_size
/2 - the_hash_algo
->rawsz
;
2511 max_size
= DELTA_SIZE(trg_entry
);
2512 ref_depth
= trg
->depth
;
2514 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
2515 (max_depth
- ref_depth
+ 1);
2518 src_size
= SIZE(src_entry
);
2519 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
2520 if (sizediff
>= max_size
)
2522 if (trg_size
< src_size
/ 32)
2525 if (!in_same_island(&trg
->entry
->idx
.oid
, &src
->entry
->idx
.oid
))
2528 /* Load data if not already done */
2530 packing_data_lock(&to_pack
);
2531 trg
->data
= read_object_file(&trg_entry
->idx
.oid
, &type
, &sz
);
2532 packing_data_unlock(&to_pack
);
2534 die(_("object %s cannot be read"),
2535 oid_to_hex(&trg_entry
->idx
.oid
));
2537 die(_("object %s inconsistent object length (%"PRIuMAX
" vs %"PRIuMAX
")"),
2538 oid_to_hex(&trg_entry
->idx
.oid
), (uintmax_t)sz
,
2539 (uintmax_t)trg_size
);
2543 packing_data_lock(&to_pack
);
2544 src
->data
= read_object_file(&src_entry
->idx
.oid
, &type
, &sz
);
2545 packing_data_unlock(&to_pack
);
2547 if (src_entry
->preferred_base
) {
2548 static int warned
= 0;
2550 warning(_("object %s cannot be read"),
2551 oid_to_hex(&src_entry
->idx
.oid
));
2553 * Those objects are not included in the
2554 * resulting pack. Be resilient and ignore
2555 * them if they can't be read, in case the
2556 * pack could be created nevertheless.
2560 die(_("object %s cannot be read"),
2561 oid_to_hex(&src_entry
->idx
.oid
));
2564 die(_("object %s inconsistent object length (%"PRIuMAX
" vs %"PRIuMAX
")"),
2565 oid_to_hex(&src_entry
->idx
.oid
), (uintmax_t)sz
,
2566 (uintmax_t)src_size
);
2570 src
->index
= create_delta_index(src
->data
, src_size
);
2572 static int warned
= 0;
2574 warning(_("suboptimal pack - out of memory"));
2577 *mem_usage
+= sizeof_delta_index(src
->index
);
2580 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
2584 if (DELTA(trg_entry
)) {
2585 /* Prefer only shallower same-sized deltas. */
2586 if (delta_size
== DELTA_SIZE(trg_entry
) &&
2587 src
->depth
+ 1 >= trg
->depth
) {
2594 * Handle memory allocation outside of the cache
2595 * accounting lock. Compiler will optimize the strangeness
2596 * away when NO_PTHREADS is defined.
2598 free(trg_entry
->delta_data
);
2600 if (trg_entry
->delta_data
) {
2601 delta_cache_size
-= DELTA_SIZE(trg_entry
);
2602 trg_entry
->delta_data
= NULL
;
2604 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
2605 delta_cache_size
+= delta_size
;
2607 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
2613 SET_DELTA(trg_entry
, src_entry
);
2614 SET_DELTA_SIZE(trg_entry
, delta_size
);
2615 trg
->depth
= src
->depth
+ 1;
2620 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
2622 struct object_entry
*child
= DELTA_CHILD(me
);
2625 const unsigned int c
= check_delta_limit(child
, n
+ 1);
2628 child
= DELTA_SIBLING(child
);
2633 static unsigned long free_unpacked(struct unpacked
*n
)
2635 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
2636 free_delta_index(n
->index
);
2639 freed_mem
+= SIZE(n
->entry
);
2640 FREE_AND_NULL(n
->data
);
2647 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
2648 int window
, int depth
, unsigned *processed
)
2650 uint32_t i
, idx
= 0, count
= 0;
2651 struct unpacked
*array
;
2652 unsigned long mem_usage
= 0;
2654 CALLOC_ARRAY(array
, window
);
2657 struct object_entry
*entry
;
2658 struct unpacked
*n
= array
+ idx
;
2659 int j
, max_depth
, best_base
= -1;
2668 if (!entry
->preferred_base
) {
2670 display_progress(progress_state
, *processed
);
2674 mem_usage
-= free_unpacked(n
);
2677 while (window_memory_limit
&&
2678 mem_usage
> window_memory_limit
&&
2680 const uint32_t tail
= (idx
+ window
- count
) % window
;
2681 mem_usage
-= free_unpacked(array
+ tail
);
2685 /* We do not compute delta to *create* objects we are not
2688 if (entry
->preferred_base
)
2692 * If the current object is at pack edge, take the depth the
2693 * objects that depend on the current object into account
2694 * otherwise they would become too deep.
2697 if (DELTA_CHILD(entry
)) {
2698 max_depth
-= check_delta_limit(entry
, 0);
2706 uint32_t other_idx
= idx
+ j
;
2708 if (other_idx
>= window
)
2709 other_idx
-= window
;
2710 m
= array
+ other_idx
;
2713 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
2717 best_base
= other_idx
;
2721 * If we decided to cache the delta data, then it is best
2722 * to compress it right away. First because we have to do
2723 * it anyway, and doing it here while we're threaded will
2724 * save a lot of time in the non threaded write phase,
2725 * as well as allow for caching more deltas within
2726 * the same cache size limit.
2728 * But only if not writing to stdout, since in that case
2729 * the network is most likely throttling writes anyway,
2730 * and therefore it is best to go to the write phase ASAP
2731 * instead, as we can afford spending more time compressing
2732 * between writes at that moment.
2734 if (entry
->delta_data
&& !pack_to_stdout
) {
2737 size
= do_compress(&entry
->delta_data
, DELTA_SIZE(entry
));
2738 if (size
< (1U << OE_Z_DELTA_BITS
)) {
2739 entry
->z_delta_size
= size
;
2741 delta_cache_size
-= DELTA_SIZE(entry
);
2742 delta_cache_size
+= entry
->z_delta_size
;
2745 FREE_AND_NULL(entry
->delta_data
);
2746 entry
->z_delta_size
= 0;
2750 /* if we made n a delta, and if n is already at max
2751 * depth, leaving it in the window is pointless. we
2752 * should evict it first.
2754 if (DELTA(entry
) && max_depth
<= n
->depth
)
2758 * Move the best delta base up in the window, after the
2759 * currently deltified object, to keep it longer. It will
2760 * be the first base object to be attempted next.
2763 struct unpacked swap
= array
[best_base
];
2764 int dist
= (window
+ idx
- best_base
) % window
;
2765 int dst
= best_base
;
2767 int src
= (dst
+ 1) % window
;
2768 array
[dst
] = array
[src
];
2776 if (count
+ 1 < window
)
2782 for (i
= 0; i
< window
; ++i
) {
2783 free_delta_index(array
[i
].index
);
2784 free(array
[i
].data
);
2790 * The main object list is split into smaller lists, each is handed to
2793 * The main thread waits on the condition that (at least) one of the workers
2794 * has stopped working (which is indicated in the .working member of
2795 * struct thread_params).
2797 * When a work thread has completed its work, it sets .working to 0 and
2798 * signals the main thread and waits on the condition that .data_ready
2801 * The main thread steals half of the work from the worker that has
2802 * most work left to hand it to the idle worker.
2805 struct thread_params
{
2807 struct object_entry
**list
;
2814 pthread_mutex_t mutex
;
2815 pthread_cond_t cond
;
2816 unsigned *processed
;
2819 static pthread_cond_t progress_cond
;
2822 * Mutex and conditional variable can't be statically-initialized on Windows.
2824 static void init_threaded_search(void)
2826 pthread_mutex_init(&cache_mutex
, NULL
);
2827 pthread_mutex_init(&progress_mutex
, NULL
);
2828 pthread_cond_init(&progress_cond
, NULL
);
2831 static void cleanup_threaded_search(void)
2833 pthread_cond_destroy(&progress_cond
);
2834 pthread_mutex_destroy(&cache_mutex
);
2835 pthread_mutex_destroy(&progress_mutex
);
2838 static void *threaded_find_deltas(void *arg
)
2840 struct thread_params
*me
= arg
;
2843 while (me
->remaining
) {
2846 find_deltas(me
->list
, &me
->remaining
,
2847 me
->window
, me
->depth
, me
->processed
);
2851 pthread_cond_signal(&progress_cond
);
2855 * We must not set ->data_ready before we wait on the
2856 * condition because the main thread may have set it to 1
2857 * before we get here. In order to be sure that new
2858 * work is available if we see 1 in ->data_ready, it
2859 * was initialized to 0 before this thread was spawned
2860 * and we reset it to 0 right away.
2862 pthread_mutex_lock(&me
->mutex
);
2863 while (!me
->data_ready
)
2864 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2866 pthread_mutex_unlock(&me
->mutex
);
2871 /* leave ->working 1 so that this doesn't get more work assigned */
2875 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2876 int window
, int depth
, unsigned *processed
)
2878 struct thread_params
*p
;
2879 int i
, ret
, active_threads
= 0;
2881 init_threaded_search();
2883 if (delta_search_threads
<= 1) {
2884 find_deltas(list
, &list_size
, window
, depth
, processed
);
2885 cleanup_threaded_search();
2888 if (progress
> pack_to_stdout
)
2889 fprintf_ln(stderr
, _("Delta compression using up to %d threads"),
2890 delta_search_threads
);
2891 CALLOC_ARRAY(p
, delta_search_threads
);
2893 /* Partition the work amongst work threads. */
2894 for (i
= 0; i
< delta_search_threads
; i
++) {
2895 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2897 /* don't use too small segments or no deltas will be found */
2898 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2901 p
[i
].window
= window
;
2903 p
[i
].processed
= processed
;
2905 p
[i
].data_ready
= 0;
2907 /* try to split chunks on "path" boundaries */
2908 while (sub_size
&& sub_size
< list_size
&&
2909 list
[sub_size
]->hash
&&
2910 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2914 p
[i
].list_size
= sub_size
;
2915 p
[i
].remaining
= sub_size
;
2918 list_size
-= sub_size
;
2921 /* Start work threads. */
2922 for (i
= 0; i
< delta_search_threads
; i
++) {
2923 if (!p
[i
].list_size
)
2925 pthread_mutex_init(&p
[i
].mutex
, NULL
);
2926 pthread_cond_init(&p
[i
].cond
, NULL
);
2927 ret
= pthread_create(&p
[i
].thread
, NULL
,
2928 threaded_find_deltas
, &p
[i
]);
2930 die(_("unable to create thread: %s"), strerror(ret
));
2935 * Now let's wait for work completion. Each time a thread is done
2936 * with its work, we steal half of the remaining work from the
2937 * thread with the largest number of unprocessed objects and give
2938 * it to that newly idle thread. This ensure good load balancing
2939 * until the remaining object list segments are simply too short
2940 * to be worth splitting anymore.
2942 while (active_threads
) {
2943 struct thread_params
*target
= NULL
;
2944 struct thread_params
*victim
= NULL
;
2945 unsigned sub_size
= 0;
2949 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
2954 pthread_cond_wait(&progress_cond
, &progress_mutex
);
2957 for (i
= 0; i
< delta_search_threads
; i
++)
2958 if (p
[i
].remaining
> 2*window
&&
2959 (!victim
|| victim
->remaining
< p
[i
].remaining
))
2962 sub_size
= victim
->remaining
/ 2;
2963 list
= victim
->list
+ victim
->list_size
- sub_size
;
2964 while (sub_size
&& list
[0]->hash
&&
2965 list
[0]->hash
== list
[-1]->hash
) {
2971 * It is possible for some "paths" to have
2972 * so many objects that no hash boundary
2973 * might be found. Let's just steal the
2974 * exact half in that case.
2976 sub_size
= victim
->remaining
/ 2;
2979 target
->list
= list
;
2980 victim
->list_size
-= sub_size
;
2981 victim
->remaining
-= sub_size
;
2983 target
->list_size
= sub_size
;
2984 target
->remaining
= sub_size
;
2985 target
->working
= 1;
2988 pthread_mutex_lock(&target
->mutex
);
2989 target
->data_ready
= 1;
2990 pthread_cond_signal(&target
->cond
);
2991 pthread_mutex_unlock(&target
->mutex
);
2994 pthread_join(target
->thread
, NULL
);
2995 pthread_cond_destroy(&target
->cond
);
2996 pthread_mutex_destroy(&target
->mutex
);
3000 cleanup_threaded_search();
3004 static int obj_is_packed(const struct object_id
*oid
)
3006 return packlist_find(&to_pack
, oid
) ||
3007 (reuse_packfile_bitmap
&&
3008 bitmap_walk_contains(bitmap_git
, reuse_packfile_bitmap
, oid
));
3011 static void add_tag_chain(const struct object_id
*oid
)
3016 * We catch duplicates already in add_object_entry(), but we'd
3017 * prefer to do this extra check to avoid having to parse the
3018 * tag at all if we already know that it's being packed (e.g., if
3019 * it was included via bitmaps, we would not have parsed it
3022 if (obj_is_packed(oid
))
3025 tag
= lookup_tag(the_repository
, oid
);
3027 if (!tag
|| parse_tag(tag
) || !tag
->tagged
)
3028 die(_("unable to pack objects reachable from tag %s"),
3031 add_object_entry(&tag
->object
.oid
, OBJ_TAG
, NULL
, 0);
3033 if (tag
->tagged
->type
!= OBJ_TAG
)
3036 tag
= (struct tag
*)tag
->tagged
;
3040 static int add_ref_tag(const char *tag UNUSED
, const struct object_id
*oid
,
3041 int flag UNUSED
, void *cb_data UNUSED
)
3043 struct object_id peeled
;
3045 if (!peel_iterated_oid(oid
, &peeled
) && obj_is_packed(&peeled
))
3050 static void prepare_pack(int window
, int depth
)
3052 struct object_entry
**delta_list
;
3053 uint32_t i
, nr_deltas
;
3056 if (use_delta_islands
)
3057 resolve_tree_islands(the_repository
, progress
, &to_pack
);
3059 get_object_details();
3062 * If we're locally repacking then we need to be doubly careful
3063 * from now on in order to make sure no stealth corruption gets
3064 * propagated to the new pack. Clients receiving streamed packs
3065 * should validate everything they get anyway so no need to incur
3066 * the additional cost here in that case.
3068 if (!pack_to_stdout
)
3069 do_check_packed_object_crc
= 1;
3071 if (!to_pack
.nr_objects
|| !window
|| !depth
)
3074 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
3077 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
3078 struct object_entry
*entry
= to_pack
.objects
+ i
;
3081 /* This happens if we decided to reuse existing
3082 * delta from a pack. "reuse_delta &&" is implied.
3086 if (!entry
->type_valid
||
3087 oe_size_less_than(&to_pack
, entry
, 50))
3090 if (entry
->no_try_delta
)
3093 if (!entry
->preferred_base
) {
3095 if (oe_type(entry
) < 0)
3096 die(_("unable to get type of object %s"),
3097 oid_to_hex(&entry
->idx
.oid
));
3099 if (oe_type(entry
) < 0) {
3101 * This object is not found, but we
3102 * don't have to include it anyway.
3108 delta_list
[n
++] = entry
;
3111 if (nr_deltas
&& n
> 1) {
3112 unsigned nr_done
= 0;
3115 progress_state
= start_progress(_("Compressing objects"),
3117 QSORT(delta_list
, n
, type_size_sort
);
3118 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
3119 stop_progress(&progress_state
);
3120 if (nr_done
!= nr_deltas
)
3121 die(_("inconsistency with delta count"));
3126 static int git_pack_config(const char *k
, const char *v
, void *cb
)
3128 if (!strcmp(k
, "pack.window")) {
3129 window
= git_config_int(k
, v
);
3132 if (!strcmp(k
, "pack.windowmemory")) {
3133 window_memory_limit
= git_config_ulong(k
, v
);
3136 if (!strcmp(k
, "pack.depth")) {
3137 depth
= git_config_int(k
, v
);
3140 if (!strcmp(k
, "pack.deltacachesize")) {
3141 max_delta_cache_size
= git_config_int(k
, v
);
3144 if (!strcmp(k
, "pack.deltacachelimit")) {
3145 cache_max_small_delta_size
= git_config_int(k
, v
);
3148 if (!strcmp(k
, "pack.writebitmaphashcache")) {
3149 if (git_config_bool(k
, v
))
3150 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
3152 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
3155 if (!strcmp(k
, "pack.writebitmaplookuptable")) {
3156 if (git_config_bool(k
, v
))
3157 write_bitmap_options
|= BITMAP_OPT_LOOKUP_TABLE
;
3159 write_bitmap_options
&= ~BITMAP_OPT_LOOKUP_TABLE
;
3162 if (!strcmp(k
, "pack.usebitmaps")) {
3163 use_bitmap_index_default
= git_config_bool(k
, v
);
3166 if (!strcmp(k
, "pack.allowpackreuse")) {
3167 allow_pack_reuse
= git_config_bool(k
, v
);
3170 if (!strcmp(k
, "pack.threads")) {
3171 delta_search_threads
= git_config_int(k
, v
);
3172 if (delta_search_threads
< 0)
3173 die(_("invalid number of threads specified (%d)"),
3174 delta_search_threads
);
3175 if (!HAVE_THREADS
&& delta_search_threads
!= 1) {
3176 warning(_("no threads support, ignoring %s"), k
);
3177 delta_search_threads
= 0;
3181 if (!strcmp(k
, "pack.indexversion")) {
3182 pack_idx_opts
.version
= git_config_int(k
, v
);
3183 if (pack_idx_opts
.version
> 2)
3184 die(_("bad pack.indexVersion=%"PRIu32
),
3185 pack_idx_opts
.version
);
3188 if (!strcmp(k
, "pack.writereverseindex")) {
3189 if (git_config_bool(k
, v
))
3190 pack_idx_opts
.flags
|= WRITE_REV
;
3192 pack_idx_opts
.flags
&= ~WRITE_REV
;
3195 if (!strcmp(k
, "uploadpack.blobpackfileuri")) {
3196 struct configured_exclusion
*ex
= xmalloc(sizeof(*ex
));
3197 const char *oid_end
, *pack_end
;
3199 * Stores the pack hash. This is not a true object ID, but is
3202 struct object_id pack_hash
;
3204 if (parse_oid_hex(v
, &ex
->e
.oid
, &oid_end
) ||
3206 parse_oid_hex(oid_end
+ 1, &pack_hash
, &pack_end
) ||
3208 die(_("value of uploadpack.blobpackfileuri must be "
3209 "of the form '<object-hash> <pack-hash> <uri>' (got '%s')"), v
);
3210 if (oidmap_get(&configured_exclusions
, &ex
->e
.oid
))
3211 die(_("object already configured in another "
3212 "uploadpack.blobpackfileuri (got '%s')"), v
);
3213 ex
->pack_hash_hex
= xcalloc(1, pack_end
- oid_end
);
3214 memcpy(ex
->pack_hash_hex
, oid_end
+ 1, pack_end
- oid_end
- 1);
3215 ex
->uri
= xstrdup(pack_end
+ 1);
3216 oidmap_put(&configured_exclusions
, ex
);
3218 return git_default_config(k
, v
, cb
);
3221 /* Counters for trace2 output when in --stdin-packs mode. */
3222 static int stdin_packs_found_nr
;
3223 static int stdin_packs_hints_nr
;
3225 static int add_object_entry_from_pack(const struct object_id
*oid
,
3226 struct packed_git
*p
,
3231 enum object_type type
= OBJ_NONE
;
3233 display_progress(progress_state
, ++nr_seen
);
3235 if (have_duplicate_entry(oid
, 0))
3238 ofs
= nth_packed_object_offset(p
, pos
);
3239 if (!want_object_in_pack(oid
, 0, &p
, &ofs
))
3243 struct rev_info
*revs
= _data
;
3244 struct object_info oi
= OBJECT_INFO_INIT
;
3247 if (packed_object_info(the_repository
, p
, ofs
, &oi
) < 0) {
3248 die(_("could not get type of object %s in pack %s"),
3249 oid_to_hex(oid
), p
->pack_name
);
3250 } else if (type
== OBJ_COMMIT
) {
3252 * commits in included packs are used as starting points for the
3253 * subsequent revision walk
3255 add_pending_oid(revs
, NULL
, oid
, 0);
3258 stdin_packs_found_nr
++;
3261 create_object_entry(oid
, type
, 0, 0, 0, p
, ofs
);
3266 static void show_commit_pack_hint(struct commit
*commit UNUSED
,
3269 /* nothing to do; commits don't have a namehash */
3272 static void show_object_pack_hint(struct object
*object
, const char *name
,
3275 struct object_entry
*oe
= packlist_find(&to_pack
, &object
->oid
);
3280 * Our 'to_pack' list was constructed by iterating all objects packed in
3281 * included packs, and so doesn't have a non-zero hash field that you
3282 * would typically pick up during a reachability traversal.
3284 * Make a best-effort attempt to fill in the ->hash and ->no_try_delta
3285 * here using a now in order to perhaps improve the delta selection
3288 oe
->hash
= pack_name_hash(name
);
3289 oe
->no_try_delta
= name
&& no_try_delta(name
);
3291 stdin_packs_hints_nr
++;
3294 static int pack_mtime_cmp(const void *_a
, const void *_b
)
3296 struct packed_git
*a
= ((const struct string_list_item
*)_a
)->util
;
3297 struct packed_git
*b
= ((const struct string_list_item
*)_b
)->util
;
3300 * order packs by descending mtime so that objects are laid out
3301 * roughly as newest-to-oldest
3303 if (a
->mtime
< b
->mtime
)
3305 else if (b
->mtime
< a
->mtime
)
3311 static void read_packs_list_from_stdin(void)
3313 struct strbuf buf
= STRBUF_INIT
;
3314 struct string_list include_packs
= STRING_LIST_INIT_DUP
;
3315 struct string_list exclude_packs
= STRING_LIST_INIT_DUP
;
3316 struct string_list_item
*item
= NULL
;
3318 struct packed_git
*p
;
3319 struct rev_info revs
;
3321 repo_init_revisions(the_repository
, &revs
, NULL
);
3323 * Use a revision walk to fill in the namehash of objects in the include
3324 * packs. To save time, we'll avoid traversing through objects that are
3325 * in excluded packs.
3327 * That may cause us to avoid populating all of the namehash fields of
3328 * all included objects, but our goal is best-effort, since this is only
3329 * an optimization during delta selection.
3331 revs
.no_kept_objects
= 1;
3332 revs
.keep_pack_cache_flags
|= IN_CORE_KEEP_PACKS
;
3333 revs
.blob_objects
= 1;
3334 revs
.tree_objects
= 1;
3335 revs
.tag_objects
= 1;
3336 revs
.ignore_missing_links
= 1;
3338 while (strbuf_getline(&buf
, stdin
) != EOF
) {
3342 if (*buf
.buf
== '^')
3343 string_list_append(&exclude_packs
, buf
.buf
+ 1);
3345 string_list_append(&include_packs
, buf
.buf
);
3350 string_list_sort(&include_packs
);
3351 string_list_sort(&exclude_packs
);
3353 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3354 const char *pack_name
= pack_basename(p
);
3356 item
= string_list_lookup(&include_packs
, pack_name
);
3358 item
= string_list_lookup(&exclude_packs
, pack_name
);
3365 * Arguments we got on stdin may not even be packs. First
3366 * check that to avoid segfaulting later on in
3367 * e.g. pack_mtime_cmp(), excluded packs are handled below.
3369 * Since we first parsed our STDIN and then sorted the input
3370 * lines the pack we error on will be whatever line happens to
3371 * sort first. This is lazy, it's enough that we report one
3372 * bad case here, we don't need to report the first/last one,
3375 for_each_string_list_item(item
, &include_packs
) {
3376 struct packed_git
*p
= item
->util
;
3378 die(_("could not find pack '%s'"), item
->string
);
3379 if (!is_pack_valid(p
))
3380 die(_("packfile %s cannot be accessed"), p
->pack_name
);
3384 * Then, handle all of the excluded packs, marking them as
3385 * kept in-core so that later calls to add_object_entry()
3386 * discards any objects that are also found in excluded packs.
3388 for_each_string_list_item(item
, &exclude_packs
) {
3389 struct packed_git
*p
= item
->util
;
3391 die(_("could not find pack '%s'"), item
->string
);
3392 p
->pack_keep_in_core
= 1;
3396 * Order packs by ascending mtime; use QSORT directly to access the
3397 * string_list_item's ->util pointer, which string_list_sort() does not
3400 QSORT(include_packs
.items
, include_packs
.nr
, pack_mtime_cmp
);
3402 for_each_string_list_item(item
, &include_packs
) {
3403 struct packed_git
*p
= item
->util
;
3404 for_each_object_in_pack(p
,
3405 add_object_entry_from_pack
,
3407 FOR_EACH_OBJECT_PACK_ORDER
);
3410 if (prepare_revision_walk(&revs
))
3411 die(_("revision walk setup failed"));
3412 traverse_commit_list(&revs
,
3413 show_commit_pack_hint
,
3414 show_object_pack_hint
,
3417 trace2_data_intmax("pack-objects", the_repository
, "stdin_packs_found",
3418 stdin_packs_found_nr
);
3419 trace2_data_intmax("pack-objects", the_repository
, "stdin_packs_hints",
3420 stdin_packs_hints_nr
);
3422 strbuf_release(&buf
);
3423 string_list_clear(&include_packs
, 0);
3424 string_list_clear(&exclude_packs
, 0);
3427 static void add_cruft_object_entry(const struct object_id
*oid
, enum object_type type
,
3428 struct packed_git
*pack
, off_t offset
,
3429 const char *name
, uint32_t mtime
)
3431 struct object_entry
*entry
;
3433 display_progress(progress_state
, ++nr_seen
);
3435 entry
= packlist_find(&to_pack
, oid
);
3438 entry
->hash
= pack_name_hash(name
);
3439 entry
->no_try_delta
= no_try_delta(name
);
3442 if (!want_object_in_pack(oid
, 0, &pack
, &offset
))
3444 if (!pack
&& type
== OBJ_BLOB
&& !has_loose_object(oid
)) {
3446 * If a traversed tree has a missing blob then we want
3447 * to avoid adding that missing object to our pack.
3449 * This only applies to missing blobs, not trees,
3450 * because the traversal needs to parse sub-trees but
3453 * Note we only perform this check when we couldn't
3454 * already find the object in a pack, so we're really
3455 * limited to "ensure non-tip blobs which don't exist in
3456 * packs do exist via loose objects". Confused?
3461 entry
= create_object_entry(oid
, type
, pack_name_hash(name
),
3462 0, name
&& no_try_delta(name
),
3466 if (mtime
> oe_cruft_mtime(&to_pack
, entry
))
3467 oe_set_cruft_mtime(&to_pack
, entry
, mtime
);
3471 static void show_cruft_object(struct object
*obj
, const char *name
, void *data UNUSED
)
3474 * if we did not record it earlier, it's at least as old as our
3475 * expiration value. Rather than find it exactly, just use that
3476 * value. This may bump it forward from its real mtime, but it
3477 * will still be "too old" next time we run with the same
3480 * if obj does appear in the packing list, this call is a noop (or may
3481 * set the namehash).
3483 add_cruft_object_entry(&obj
->oid
, obj
->type
, NULL
, 0, name
, cruft_expiration
);
3486 static void show_cruft_commit(struct commit
*commit
, void *data
)
3488 show_cruft_object((struct object
*)commit
, NULL
, data
);
3491 static int cruft_include_check_obj(struct object
*obj
, void *data UNUSED
)
3493 return !has_object_kept_pack(&obj
->oid
, IN_CORE_KEEP_PACKS
);
3496 static int cruft_include_check(struct commit
*commit
, void *data
)
3498 return cruft_include_check_obj((struct object
*)commit
, data
);
3501 static void set_cruft_mtime(const struct object
*object
,
3502 struct packed_git
*pack
,
3503 off_t offset
, time_t mtime
)
3505 add_cruft_object_entry(&object
->oid
, object
->type
, pack
, offset
, NULL
,
3509 static void mark_pack_kept_in_core(struct string_list
*packs
, unsigned keep
)
3511 struct string_list_item
*item
= NULL
;
3512 for_each_string_list_item(item
, packs
) {
3513 struct packed_git
*p
= item
->util
;
3515 die(_("could not find pack '%s'"), item
->string
);
3516 p
->pack_keep_in_core
= keep
;
3520 static void add_unreachable_loose_objects(void);
3521 static void add_objects_in_unpacked_packs(void);
3523 static void enumerate_cruft_objects(void)
3526 progress_state
= start_progress(_("Enumerating cruft objects"), 0);
3528 add_objects_in_unpacked_packs();
3529 add_unreachable_loose_objects();
3531 stop_progress(&progress_state
);
3534 static void enumerate_and_traverse_cruft_objects(struct string_list
*fresh_packs
)
3536 struct packed_git
*p
;
3537 struct rev_info revs
;
3540 repo_init_revisions(the_repository
, &revs
, NULL
);
3542 revs
.tag_objects
= 1;
3543 revs
.tree_objects
= 1;
3544 revs
.blob_objects
= 1;
3546 revs
.include_check
= cruft_include_check
;
3547 revs
.include_check_obj
= cruft_include_check_obj
;
3549 revs
.ignore_missing_links
= 1;
3552 progress_state
= start_progress(_("Enumerating cruft objects"), 0);
3553 ret
= add_unseen_recent_objects_to_traversal(&revs
, cruft_expiration
,
3554 set_cruft_mtime
, 1);
3555 stop_progress(&progress_state
);
3558 die(_("unable to add cruft objects"));
3561 * Re-mark only the fresh packs as kept so that objects in
3562 * unknown packs do not halt the reachability traversal early.
3564 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
)
3565 p
->pack_keep_in_core
= 0;
3566 mark_pack_kept_in_core(fresh_packs
, 1);
3568 if (prepare_revision_walk(&revs
))
3569 die(_("revision walk setup failed"));
3571 progress_state
= start_progress(_("Traversing cruft objects"), 0);
3573 traverse_commit_list(&revs
, show_cruft_commit
, show_cruft_object
, NULL
);
3575 stop_progress(&progress_state
);
3578 static void read_cruft_objects(void)
3580 struct strbuf buf
= STRBUF_INIT
;
3581 struct string_list discard_packs
= STRING_LIST_INIT_DUP
;
3582 struct string_list fresh_packs
= STRING_LIST_INIT_DUP
;
3583 struct packed_git
*p
;
3585 ignore_packed_keep_in_core
= 1;
3587 while (strbuf_getline(&buf
, stdin
) != EOF
) {
3591 if (*buf
.buf
== '-')
3592 string_list_append(&discard_packs
, buf
.buf
+ 1);
3594 string_list_append(&fresh_packs
, buf
.buf
);
3598 string_list_sort(&discard_packs
);
3599 string_list_sort(&fresh_packs
);
3601 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3602 const char *pack_name
= pack_basename(p
);
3603 struct string_list_item
*item
;
3605 item
= string_list_lookup(&fresh_packs
, pack_name
);
3607 item
= string_list_lookup(&discard_packs
, pack_name
);
3613 * This pack wasn't mentioned in either the "fresh" or
3614 * "discard" list, so the caller didn't know about it.
3616 * Mark it as kept so that its objects are ignored by
3617 * add_unseen_recent_objects_to_traversal(). We'll
3618 * unmark it before starting the traversal so it doesn't
3619 * halt the traversal early.
3621 p
->pack_keep_in_core
= 1;
3625 mark_pack_kept_in_core(&fresh_packs
, 1);
3626 mark_pack_kept_in_core(&discard_packs
, 0);
3628 if (cruft_expiration
)
3629 enumerate_and_traverse_cruft_objects(&fresh_packs
);
3631 enumerate_cruft_objects();
3633 strbuf_release(&buf
);
3634 string_list_clear(&discard_packs
, 0);
3635 string_list_clear(&fresh_packs
, 0);
3638 static void read_object_list_from_stdin(void)
3640 char line
[GIT_MAX_HEXSZ
+ 1 + PATH_MAX
+ 2];
3641 struct object_id oid
;
3645 if (!fgets(line
, sizeof(line
), stdin
)) {
3649 BUG("fgets returned NULL, not EOF, not error!");
3655 if (line
[0] == '-') {
3656 if (get_oid_hex(line
+1, &oid
))
3657 die(_("expected edge object ID, got garbage:\n %s"),
3659 add_preferred_base(&oid
);
3662 if (parse_oid_hex(line
, &oid
, &p
))
3663 die(_("expected object ID, got garbage:\n %s"), line
);
3665 add_preferred_base_object(p
+ 1);
3666 add_object_entry(&oid
, OBJ_NONE
, p
+ 1, 0);
3670 static void show_commit(struct commit
*commit
, void *data UNUSED
)
3672 add_object_entry(&commit
->object
.oid
, OBJ_COMMIT
, NULL
, 0);
3674 if (write_bitmap_index
)
3675 index_commit_for_bitmap(commit
);
3677 if (use_delta_islands
)
3678 propagate_island_marks(commit
);
3681 static void show_object(struct object
*obj
, const char *name
,
3684 add_preferred_base_object(name
);
3685 add_object_entry(&obj
->oid
, obj
->type
, name
, 0);
3687 if (use_delta_islands
) {
3690 struct object_entry
*ent
;
3692 /* the empty string is a root tree, which is depth 0 */
3693 depth
= *name
? 1 : 0;
3694 for (p
= strchr(name
, '/'); p
; p
= strchr(p
+ 1, '/'))
3697 ent
= packlist_find(&to_pack
, &obj
->oid
);
3698 if (ent
&& depth
> oe_tree_depth(&to_pack
, ent
))
3699 oe_set_tree_depth(&to_pack
, ent
, depth
);
3703 static void show_object__ma_allow_any(struct object
*obj
, const char *name
, void *data
)
3705 assert(arg_missing_action
== MA_ALLOW_ANY
);
3708 * Quietly ignore ALL missing objects. This avoids problems with
3709 * staging them now and getting an odd error later.
3711 if (!has_object(the_repository
, &obj
->oid
, 0))
3714 show_object(obj
, name
, data
);
3717 static void show_object__ma_allow_promisor(struct object
*obj
, const char *name
, void *data
)
3719 assert(arg_missing_action
== MA_ALLOW_PROMISOR
);
3722 * Quietly ignore EXPECTED missing objects. This avoids problems with
3723 * staging them now and getting an odd error later.
3725 if (!has_object(the_repository
, &obj
->oid
, 0) && is_promisor_object(&obj
->oid
))
3728 show_object(obj
, name
, data
);
3731 static int option_parse_missing_action(const struct option
*opt
,
3732 const char *arg
, int unset
)
3737 if (!strcmp(arg
, "error")) {
3738 arg_missing_action
= MA_ERROR
;
3739 fn_show_object
= show_object
;
3743 if (!strcmp(arg
, "allow-any")) {
3744 arg_missing_action
= MA_ALLOW_ANY
;
3745 fetch_if_missing
= 0;
3746 fn_show_object
= show_object__ma_allow_any
;
3750 if (!strcmp(arg
, "allow-promisor")) {
3751 arg_missing_action
= MA_ALLOW_PROMISOR
;
3752 fetch_if_missing
= 0;
3753 fn_show_object
= show_object__ma_allow_promisor
;
3757 die(_("invalid value for '%s': '%s'"), "--missing", arg
);
3761 static void show_edge(struct commit
*commit
)
3763 add_preferred_base(&commit
->object
.oid
);
3766 static int add_object_in_unpacked_pack(const struct object_id
*oid
,
3767 struct packed_git
*pack
,
3775 if (pack
->is_cruft
) {
3776 if (load_pack_mtimes(pack
) < 0)
3777 die(_("could not load cruft pack .mtimes"));
3778 mtime
= nth_packed_mtime(pack
, pos
);
3780 mtime
= pack
->mtime
;
3782 offset
= nth_packed_object_offset(pack
, pos
);
3784 add_cruft_object_entry(oid
, OBJ_NONE
, pack
, offset
,
3787 add_object_entry(oid
, OBJ_NONE
, "", 0);
3792 static void add_objects_in_unpacked_packs(void)
3794 if (for_each_packed_object(add_object_in_unpacked_pack
, NULL
,
3795 FOR_EACH_OBJECT_PACK_ORDER
|
3796 FOR_EACH_OBJECT_LOCAL_ONLY
|
3797 FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS
|
3798 FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS
))
3799 die(_("cannot open pack index"));
3802 static int add_loose_object(const struct object_id
*oid
, const char *path
,
3805 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
3808 warning(_("loose object at %s could not be examined"), path
);
3814 if (stat(path
, &st
) < 0) {
3815 if (errno
== ENOENT
)
3817 return error_errno("unable to stat %s", oid_to_hex(oid
));
3820 add_cruft_object_entry(oid
, type
, NULL
, 0, NULL
,
3823 add_object_entry(oid
, type
, "", 0);
3829 * We actually don't even have to worry about reachability here.
3830 * add_object_entry will weed out duplicates, so we just add every
3831 * loose object we find.
3833 static void add_unreachable_loose_objects(void)
3835 for_each_loose_file_in_objdir(get_object_directory(),
3840 static int has_sha1_pack_kept_or_nonlocal(const struct object_id
*oid
)
3842 static struct packed_git
*last_found
= (void *)1;
3843 struct packed_git
*p
;
3845 p
= (last_found
!= (void *)1) ? last_found
:
3846 get_all_packs(the_repository
);
3849 if ((!p
->pack_local
|| p
->pack_keep
||
3850 p
->pack_keep_in_core
) &&
3851 find_pack_entry_one(oid
->hash
, p
)) {
3855 if (p
== last_found
)
3856 p
= get_all_packs(the_repository
);
3859 if (p
== last_found
)
3866 * Store a list of sha1s that are should not be discarded
3867 * because they are either written too recently, or are
3868 * reachable from another object that was.
3870 * This is filled by get_object_list.
3872 static struct oid_array recent_objects
;
3874 static int loosened_object_can_be_discarded(const struct object_id
*oid
,
3877 if (!unpack_unreachable_expiration
)
3879 if (mtime
> unpack_unreachable_expiration
)
3881 if (oid_array_lookup(&recent_objects
, oid
) >= 0)
3886 static void loosen_unused_packed_objects(void)
3888 struct packed_git
*p
;
3890 uint32_t loosened_objects_nr
= 0;
3891 struct object_id oid
;
3893 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
3894 if (!p
->pack_local
|| p
->pack_keep
|| p
->pack_keep_in_core
)
3897 if (open_pack_index(p
))
3898 die(_("cannot open pack index"));
3900 for (i
= 0; i
< p
->num_objects
; i
++) {
3901 nth_packed_object_id(&oid
, p
, i
);
3902 if (!packlist_find(&to_pack
, &oid
) &&
3903 !has_sha1_pack_kept_or_nonlocal(&oid
) &&
3904 !loosened_object_can_be_discarded(&oid
, p
->mtime
)) {
3905 if (force_object_loose(&oid
, p
->mtime
))
3906 die(_("unable to force loose object"));
3907 loosened_objects_nr
++;
3912 trace2_data_intmax("pack-objects", the_repository
,
3913 "loosen_unused_packed_objects/loosened", loosened_objects_nr
);
3917 * This tracks any options which pack-reuse code expects to be on, or which a
3918 * reader of the pack might not understand, and which would therefore prevent
3919 * blind reuse of what we have on disk.
3921 static int pack_options_allow_reuse(void)
3923 return allow_pack_reuse
&&
3925 !ignore_packed_keep_on_disk
&&
3926 !ignore_packed_keep_in_core
&&
3927 (!local
|| !have_non_local_packs
) &&
3931 static int get_object_list_from_bitmap(struct rev_info
*revs
)
3933 if (!(bitmap_git
= prepare_bitmap_walk(revs
, 0)))
3936 if (pack_options_allow_reuse() &&
3937 !reuse_partial_packfile_from_bitmap(
3940 &reuse_packfile_objects
,
3941 &reuse_packfile_bitmap
)) {
3942 assert(reuse_packfile_objects
);
3943 nr_result
+= reuse_packfile_objects
;
3944 nr_seen
+= reuse_packfile_objects
;
3945 display_progress(progress_state
, nr_seen
);
3948 traverse_bitmap_commit_list(bitmap_git
, revs
,
3949 &add_object_entry_from_bitmap
);
3953 static void record_recent_object(struct object
*obj
,
3954 const char *name UNUSED
,
3957 oid_array_append(&recent_objects
, &obj
->oid
);
3960 static void record_recent_commit(struct commit
*commit
, void *data UNUSED
)
3962 oid_array_append(&recent_objects
, &commit
->object
.oid
);
3965 static int mark_bitmap_preferred_tip(const char *refname
,
3966 const struct object_id
*oid
,
3970 struct object_id peeled
;
3971 struct object
*object
;
3973 if (!peel_iterated_oid(oid
, &peeled
))
3976 object
= parse_object_or_die(oid
, refname
);
3977 if (object
->type
== OBJ_COMMIT
)
3978 object
->flags
|= NEEDS_BITMAP
;
3983 static void mark_bitmap_preferred_tips(void)
3985 struct string_list_item
*item
;
3986 const struct string_list
*preferred_tips
;
3988 preferred_tips
= bitmap_preferred_tips(the_repository
);
3989 if (!preferred_tips
)
3992 for_each_string_list_item(item
, preferred_tips
) {
3993 for_each_ref_in(item
->string
, mark_bitmap_preferred_tip
, NULL
);
3997 static void get_object_list(struct rev_info
*revs
, int ac
, const char **av
)
3999 struct setup_revision_opt s_r_opt
= {
4000 .allow_exclude_promisor_objects
= 1,
4006 save_commit_buffer
= 0;
4007 setup_revisions(ac
, av
, revs
, &s_r_opt
);
4009 /* make sure shallows are read */
4010 is_repository_shallow(the_repository
);
4012 save_warning
= warn_on_object_refname_ambiguity
;
4013 warn_on_object_refname_ambiguity
= 0;
4015 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
4016 int len
= strlen(line
);
4017 if (len
&& line
[len
- 1] == '\n')
4022 if (!strcmp(line
, "--not")) {
4023 flags
^= UNINTERESTING
;
4024 write_bitmap_index
= 0;
4027 if (starts_with(line
, "--shallow ")) {
4028 struct object_id oid
;
4029 if (get_oid_hex(line
+ 10, &oid
))
4030 die("not an object name '%s'", line
+ 10);
4031 register_shallow(the_repository
, &oid
);
4032 use_bitmap_index
= 0;
4035 die(_("not a rev '%s'"), line
);
4037 if (handle_revision_arg(line
, revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
4038 die(_("bad revision '%s'"), line
);
4041 warn_on_object_refname_ambiguity
= save_warning
;
4043 if (use_bitmap_index
&& !get_object_list_from_bitmap(revs
))
4046 if (use_delta_islands
)
4047 load_delta_islands(the_repository
, progress
);
4049 if (write_bitmap_index
)
4050 mark_bitmap_preferred_tips();
4052 if (prepare_revision_walk(revs
))
4053 die(_("revision walk setup failed"));
4054 mark_edges_uninteresting(revs
, show_edge
, sparse
);
4056 if (!fn_show_object
)
4057 fn_show_object
= show_object
;
4058 traverse_commit_list(revs
,
4059 show_commit
, fn_show_object
,
4062 if (unpack_unreachable_expiration
) {
4063 revs
->ignore_missing_links
= 1;
4064 if (add_unseen_recent_objects_to_traversal(revs
,
4065 unpack_unreachable_expiration
, NULL
, 0))
4066 die(_("unable to add recent objects"));
4067 if (prepare_revision_walk(revs
))
4068 die(_("revision walk setup failed"));
4069 traverse_commit_list(revs
, record_recent_commit
,
4070 record_recent_object
, NULL
);
4073 if (keep_unreachable
)
4074 add_objects_in_unpacked_packs();
4075 if (pack_loose_unreachable
)
4076 add_unreachable_loose_objects();
4077 if (unpack_unreachable
)
4078 loosen_unused_packed_objects();
4080 oid_array_clear(&recent_objects
);
4083 static void add_extra_kept_packs(const struct string_list
*names
)
4085 struct packed_git
*p
;
4090 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
4091 const char *name
= basename(p
->pack_name
);
4097 for (i
= 0; i
< names
->nr
; i
++)
4098 if (!fspathcmp(name
, names
->items
[i
].string
))
4101 if (i
< names
->nr
) {
4102 p
->pack_keep_in_core
= 1;
4103 ignore_packed_keep_in_core
= 1;
4109 static int option_parse_index_version(const struct option
*opt
,
4110 const char *arg
, int unset
)
4113 const char *val
= arg
;
4115 BUG_ON_OPT_NEG(unset
);
4117 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
4118 if (pack_idx_opts
.version
> 2)
4119 die(_("unsupported index version %s"), val
);
4120 if (*c
== ',' && c
[1])
4121 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
4122 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
4123 die(_("bad index version '%s'"), val
);
4127 static int option_parse_unpack_unreachable(const struct option
*opt
,
4128 const char *arg
, int unset
)
4131 unpack_unreachable
= 0;
4132 unpack_unreachable_expiration
= 0;
4135 unpack_unreachable
= 1;
4137 unpack_unreachable_expiration
= approxidate(arg
);
4142 static int option_parse_cruft_expiration(const struct option
*opt
,
4143 const char *arg
, int unset
)
4147 cruft_expiration
= 0;
4151 cruft_expiration
= approxidate(arg
);
4156 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
4158 int use_internal_rev_list
= 0;
4160 int all_progress_implied
= 0;
4161 struct strvec rp
= STRVEC_INIT
;
4162 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
4163 int rev_list_index
= 0;
4164 int stdin_packs
= 0;
4165 struct string_list keep_pack_list
= STRING_LIST_INIT_NODUP
;
4166 struct list_objects_filter_options filter_options
=
4167 LIST_OBJECTS_FILTER_INIT
;
4169 struct option pack_objects_options
[] = {
4170 OPT_SET_INT('q', "quiet", &progress
,
4171 N_("do not show progress meter"), 0),
4172 OPT_SET_INT(0, "progress", &progress
,
4173 N_("show progress meter"), 1),
4174 OPT_SET_INT(0, "all-progress", &progress
,
4175 N_("show progress meter during object writing phase"), 2),
4176 OPT_BOOL(0, "all-progress-implied",
4177 &all_progress_implied
,
4178 N_("similar to --all-progress when progress meter is shown")),
4179 OPT_CALLBACK_F(0, "index-version", NULL
, N_("<version>[,<offset>]"),
4180 N_("write the pack index file in the specified idx format version"),
4181 PARSE_OPT_NONEG
, option_parse_index_version
),
4182 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
4183 N_("maximum size of each output pack file")),
4184 OPT_BOOL(0, "local", &local
,
4185 N_("ignore borrowed objects from alternate object store")),
4186 OPT_BOOL(0, "incremental", &incremental
,
4187 N_("ignore packed objects")),
4188 OPT_INTEGER(0, "window", &window
,
4189 N_("limit pack window by objects")),
4190 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
4191 N_("limit pack window by memory in addition to object limit")),
4192 OPT_INTEGER(0, "depth", &depth
,
4193 N_("maximum length of delta chain allowed in the resulting pack")),
4194 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
4195 N_("reuse existing deltas")),
4196 OPT_BOOL(0, "reuse-object", &reuse_object
,
4197 N_("reuse existing objects")),
4198 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
4199 N_("use OFS_DELTA objects")),
4200 OPT_INTEGER(0, "threads", &delta_search_threads
,
4201 N_("use threads when searching for best delta matches")),
4202 OPT_BOOL(0, "non-empty", &non_empty
,
4203 N_("do not create an empty pack output")),
4204 OPT_BOOL(0, "revs", &use_internal_rev_list
,
4205 N_("read revision arguments from standard input")),
4206 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked
,
4207 N_("limit the objects to those that are not yet packed"),
4208 1, PARSE_OPT_NONEG
),
4209 OPT_SET_INT_F(0, "all", &rev_list_all
,
4210 N_("include objects reachable from any reference"),
4211 1, PARSE_OPT_NONEG
),
4212 OPT_SET_INT_F(0, "reflog", &rev_list_reflog
,
4213 N_("include objects referred by reflog entries"),
4214 1, PARSE_OPT_NONEG
),
4215 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index
,
4216 N_("include objects referred to by the index"),
4217 1, PARSE_OPT_NONEG
),
4218 OPT_BOOL(0, "stdin-packs", &stdin_packs
,
4219 N_("read packs from stdin")),
4220 OPT_BOOL(0, "stdout", &pack_to_stdout
,
4221 N_("output pack to stdout")),
4222 OPT_BOOL(0, "include-tag", &include_tag
,
4223 N_("include tag objects that refer to objects to be packed")),
4224 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
4225 N_("keep unreachable objects")),
4226 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
4227 N_("pack loose unreachable objects")),
4228 OPT_CALLBACK_F(0, "unpack-unreachable", NULL
, N_("time"),
4229 N_("unpack unreachable objects newer than <time>"),
4230 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
),
4231 OPT_BOOL(0, "cruft", &cruft
, N_("create a cruft pack")),
4232 OPT_CALLBACK_F(0, "cruft-expiration", NULL
, N_("time"),
4233 N_("expire cruft objects older than <time>"),
4234 PARSE_OPT_OPTARG
, option_parse_cruft_expiration
),
4235 OPT_BOOL(0, "sparse", &sparse
,
4236 N_("use the sparse reachability algorithm")),
4237 OPT_BOOL(0, "thin", &thin
,
4238 N_("create thin packs")),
4239 OPT_BOOL(0, "shallow", &shallow
,
4240 N_("create packs suitable for shallow fetches")),
4241 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk
,
4242 N_("ignore packs that have companion .keep file")),
4243 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list
, N_("name"),
4244 N_("ignore this pack")),
4245 OPT_INTEGER(0, "compression", &pack_compression_level
,
4246 N_("pack compression level")),
4247 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
4248 N_("do not hide commits by grafts"), 0),
4249 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
4250 N_("use a bitmap index if available to speed up counting objects")),
4251 OPT_SET_INT(0, "write-bitmap-index", &write_bitmap_index
,
4252 N_("write a bitmap index together with the pack index"),
4254 OPT_SET_INT_F(0, "write-bitmap-index-quiet",
4255 &write_bitmap_index
,
4256 N_("write a bitmap index if possible"),
4257 WRITE_BITMAP_QUIET
, PARSE_OPT_HIDDEN
),
4258 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options
),
4259 OPT_CALLBACK_F(0, "missing", NULL
, N_("action"),
4260 N_("handling for missing objects"), PARSE_OPT_NONEG
,
4261 option_parse_missing_action
),
4262 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects
,
4263 N_("do not pack objects in promisor packfiles")),
4264 OPT_BOOL(0, "delta-islands", &use_delta_islands
,
4265 N_("respect islands during delta compression")),
4266 OPT_STRING_LIST(0, "uri-protocol", &uri_protocols
,
4268 N_("exclude any configured uploadpack.blobpackfileuri with this protocol")),
4272 if (DFS_NUM_STATES
> (1 << OE_DFS_STATE_BITS
))
4273 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
4275 read_replace_refs
= 0;
4277 sparse
= git_env_bool("GIT_TEST_PACK_SPARSE", -1);
4278 if (the_repository
->gitdir
) {
4279 prepare_repo_settings(the_repository
);
4281 sparse
= the_repository
->settings
.pack_use_sparse
;
4284 reset_pack_idx_option(&pack_idx_opts
);
4285 git_config(git_pack_config
, NULL
);
4286 if (git_env_bool(GIT_TEST_WRITE_REV_INDEX
, 0))
4287 pack_idx_opts
.flags
|= WRITE_REV
;
4289 progress
= isatty(2);
4290 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
4294 base_name
= argv
[0];
4297 if (pack_to_stdout
!= !base_name
|| argc
)
4298 usage_with_options(pack_usage
, pack_objects_options
);
4302 if (depth
>= (1 << OE_DEPTH_BITS
)) {
4303 warning(_("delta chain depth %d is too deep, forcing %d"),
4304 depth
, (1 << OE_DEPTH_BITS
) - 1);
4305 depth
= (1 << OE_DEPTH_BITS
) - 1;
4307 if (cache_max_small_delta_size
>= (1U << OE_Z_DELTA_BITS
)) {
4308 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
4309 (1U << OE_Z_DELTA_BITS
) - 1);
4310 cache_max_small_delta_size
= (1U << OE_Z_DELTA_BITS
) - 1;
4315 strvec_push(&rp
, "pack-objects");
4317 use_internal_rev_list
= 1;
4318 strvec_push(&rp
, shallow
4319 ? "--objects-edge-aggressive"
4320 : "--objects-edge");
4322 strvec_push(&rp
, "--objects");
4325 use_internal_rev_list
= 1;
4326 strvec_push(&rp
, "--all");
4328 if (rev_list_reflog
) {
4329 use_internal_rev_list
= 1;
4330 strvec_push(&rp
, "--reflog");
4332 if (rev_list_index
) {
4333 use_internal_rev_list
= 1;
4334 strvec_push(&rp
, "--indexed-objects");
4336 if (rev_list_unpacked
&& !stdin_packs
) {
4337 use_internal_rev_list
= 1;
4338 strvec_push(&rp
, "--unpacked");
4341 if (exclude_promisor_objects
) {
4342 use_internal_rev_list
= 1;
4343 fetch_if_missing
= 0;
4344 strvec_push(&rp
, "--exclude-promisor-objects");
4346 if (unpack_unreachable
|| keep_unreachable
|| pack_loose_unreachable
)
4347 use_internal_rev_list
= 1;
4351 if (pack_compression_level
== -1)
4352 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
4353 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
4354 die(_("bad pack compression level %d"), pack_compression_level
);
4356 if (!delta_search_threads
) /* --threads=0 means autodetect */
4357 delta_search_threads
= online_cpus();
4359 if (!HAVE_THREADS
&& delta_search_threads
!= 1)
4360 warning(_("no threads support, ignoring --threads"));
4361 if (!pack_to_stdout
&& !pack_size_limit
&& !cruft
)
4362 pack_size_limit
= pack_size_limit_cfg
;
4363 if (pack_to_stdout
&& pack_size_limit
)
4364 die(_("--max-pack-size cannot be used to build a pack for transfer"));
4365 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
4366 warning(_("minimum pack size limit is 1 MiB"));
4367 pack_size_limit
= 1024*1024;
4370 if (!pack_to_stdout
&& thin
)
4371 die(_("--thin cannot be used to build an indexable pack"));
4373 if (keep_unreachable
&& unpack_unreachable
)
4374 die(_("options '%s' and '%s' cannot be used together"), "--keep-unreachable", "--unpack-unreachable");
4375 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
4376 unpack_unreachable_expiration
= 0;
4378 if (filter_options
.choice
) {
4379 if (!pack_to_stdout
)
4380 die(_("cannot use --filter without --stdout"));
4382 die(_("cannot use --filter with --stdin-packs"));
4385 if (stdin_packs
&& use_internal_rev_list
)
4386 die(_("cannot use internal rev list with --stdin-packs"));
4389 if (use_internal_rev_list
)
4390 die(_("cannot use internal rev list with --cruft"));
4392 die(_("cannot use --stdin-packs with --cruft"));
4393 if (pack_size_limit
)
4394 die(_("cannot use --max-pack-size with --cruft"));
4398 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
4400 * - to produce good pack (with bitmap index not-yet-packed objects are
4401 * packed in suboptimal order).
4403 * - to use more robust pack-generation codepath (avoiding possible
4404 * bugs in bitmap code and possible bitmap index corruption).
4406 if (!pack_to_stdout
)
4407 use_bitmap_index_default
= 0;
4409 if (use_bitmap_index
< 0)
4410 use_bitmap_index
= use_bitmap_index_default
;
4412 /* "hard" reasons not to use bitmaps; these just won't work at all */
4413 if (!use_internal_rev_list
|| (!pack_to_stdout
&& write_bitmap_index
) || is_repository_shallow(the_repository
))
4414 use_bitmap_index
= 0;
4416 if (pack_to_stdout
|| !rev_list_all
)
4417 write_bitmap_index
= 0;
4419 if (use_delta_islands
)
4420 strvec_push(&rp
, "--topo-order");
4422 if (progress
&& all_progress_implied
)
4425 add_extra_kept_packs(&keep_pack_list
);
4426 if (ignore_packed_keep_on_disk
) {
4427 struct packed_git
*p
;
4428 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
)
4429 if (p
->pack_local
&& p
->pack_keep
)
4431 if (!p
) /* no keep-able packs found */
4432 ignore_packed_keep_on_disk
= 0;
4436 * unlike ignore_packed_keep_on_disk above, we do not
4437 * want to unset "local" based on looking at packs, as
4438 * it also covers non-local objects
4440 struct packed_git
*p
;
4441 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
4442 if (!p
->pack_local
) {
4443 have_non_local_packs
= 1;
4449 trace2_region_enter("pack-objects", "enumerate-objects",
4451 prepare_packing_data(the_repository
, &to_pack
);
4453 if (progress
&& !cruft
)
4454 progress_state
= start_progress(_("Enumerating objects"), 0);
4456 /* avoids adding objects in excluded packs */
4457 ignore_packed_keep_in_core
= 1;
4458 read_packs_list_from_stdin();
4459 if (rev_list_unpacked
)
4460 add_unreachable_loose_objects();
4462 read_cruft_objects();
4463 } else if (!use_internal_rev_list
) {
4464 read_object_list_from_stdin();
4466 struct rev_info revs
;
4468 repo_init_revisions(the_repository
, &revs
, NULL
);
4469 list_objects_filter_copy(&revs
.filter
, &filter_options
);
4470 get_object_list(&revs
, rp
.nr
, rp
.v
);
4471 release_revisions(&revs
);
4473 cleanup_preferred_base();
4474 if (include_tag
&& nr_result
)
4475 for_each_tag_ref(add_ref_tag
, NULL
);
4476 stop_progress(&progress_state
);
4477 trace2_region_leave("pack-objects", "enumerate-objects",
4480 if (non_empty
&& !nr_result
)
4483 trace2_region_enter("pack-objects", "prepare-pack",
4485 prepare_pack(window
, depth
);
4486 trace2_region_leave("pack-objects", "prepare-pack",
4490 trace2_region_enter("pack-objects", "write-pack-file", the_repository
);
4491 write_excluded_by_configs();
4493 trace2_region_leave("pack-objects", "write-pack-file", the_repository
);
4497 _("Total %"PRIu32
" (delta %"PRIu32
"),"
4498 " reused %"PRIu32
" (delta %"PRIu32
"),"
4499 " pack-reused %"PRIu32
),
4500 written
, written_delta
, reused
, reused_delta
,
4501 reuse_packfile_objects
);
4504 list_objects_filter_release(&filter_options
);