12 #include "pack-revindex.h"
13 #include "csum-file.h"
14 #include "tree-walk.h"
17 #include "list-objects.h"
18 #include "pack-objects.h"
21 #include "streaming.h"
22 #include "thread-utils.h"
23 #include "pack-bitmap.h"
24 #include "reachable.h"
25 #include "sha1-array.h"
26 #include "argv-array.h"
29 static const char *pack_usage
[] = {
30 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
31 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
36 * Objects we are going to pack are collected in the `to_pack` structure.
37 * It contains an array (dynamically expanded) of the object data, and a map
38 * that can resolve SHA1s to their position in the array.
40 static struct packing_data to_pack
;
42 static struct pack_idx_entry
**written_list
;
43 static uint32_t nr_result
, nr_written
;
46 static int reuse_delta
= 1, reuse_object
= 1;
47 static int keep_unreachable
, unpack_unreachable
, include_tag
;
48 static unsigned long unpack_unreachable_expiration
;
49 static int pack_loose_unreachable
;
51 static int have_non_local_packs
;
52 static int incremental
;
53 static int ignore_packed_keep
;
54 static int allow_ofs_delta
;
55 static struct pack_idx_option pack_idx_opts
;
56 static const char *base_name
;
57 static int progress
= 1;
58 static int window
= 10;
59 static unsigned long pack_size_limit
;
60 static int depth
= 50;
61 static int delta_search_threads
;
62 static int pack_to_stdout
;
63 static int num_preferred_base
;
64 static struct progress
*progress_state
;
66 static struct packed_git
*reuse_packfile
;
67 static uint32_t reuse_packfile_objects
;
68 static off_t reuse_packfile_offset
;
70 static int use_bitmap_index_default
= 1;
71 static int use_bitmap_index
= -1;
72 static int write_bitmap_index
;
73 static uint16_t write_bitmap_options
;
75 static unsigned long delta_cache_size
= 0;
76 static unsigned long max_delta_cache_size
= 256 * 1024 * 1024;
77 static unsigned long cache_max_small_delta_size
= 1000;
79 static unsigned long window_memory_limit
= 0;
84 static uint32_t written
, written_delta
;
85 static uint32_t reused
, reused_delta
;
90 static struct commit
**indexed_commits
;
91 static unsigned int indexed_commits_nr
;
92 static unsigned int indexed_commits_alloc
;
94 static void index_commit_for_bitmap(struct commit
*commit
)
96 if (indexed_commits_nr
>= indexed_commits_alloc
) {
97 indexed_commits_alloc
= (indexed_commits_alloc
+ 32) * 2;
98 REALLOC_ARRAY(indexed_commits
, indexed_commits_alloc
);
101 indexed_commits
[indexed_commits_nr
++] = commit
;
104 static void *get_delta(struct object_entry
*entry
)
106 unsigned long size
, base_size
, delta_size
;
107 void *buf
, *base_buf
, *delta_buf
;
108 enum object_type type
;
110 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
112 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
113 base_buf
= read_sha1_file(entry
->delta
->idx
.sha1
, &type
, &base_size
);
115 die("unable to read %s", sha1_to_hex(entry
->delta
->idx
.sha1
));
116 delta_buf
= diff_delta(base_buf
, base_size
,
117 buf
, size
, &delta_size
, 0);
118 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
119 die("delta size changed");
125 static unsigned long do_compress(void **pptr
, unsigned long size
)
129 unsigned long maxsize
;
131 git_deflate_init(&stream
, pack_compression_level
);
132 maxsize
= git_deflate_bound(&stream
, size
);
135 out
= xmalloc(maxsize
);
139 stream
.avail_in
= size
;
140 stream
.next_out
= out
;
141 stream
.avail_out
= maxsize
;
142 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
144 git_deflate_end(&stream
);
147 return stream
.total_out
;
150 static unsigned long write_large_blob_data(struct git_istream
*st
, struct sha1file
*f
,
151 const unsigned char *sha1
)
154 unsigned char ibuf
[1024 * 16];
155 unsigned char obuf
[1024 * 16];
156 unsigned long olen
= 0;
158 git_deflate_init(&stream
, pack_compression_level
);
163 readlen
= read_istream(st
, ibuf
, sizeof(ibuf
));
165 die(_("unable to read %s"), sha1_to_hex(sha1
));
167 stream
.next_in
= ibuf
;
168 stream
.avail_in
= readlen
;
169 while ((stream
.avail_in
|| readlen
== 0) &&
170 (zret
== Z_OK
|| zret
== Z_BUF_ERROR
)) {
171 stream
.next_out
= obuf
;
172 stream
.avail_out
= sizeof(obuf
);
173 zret
= git_deflate(&stream
, readlen
? 0 : Z_FINISH
);
174 sha1write(f
, obuf
, stream
.next_out
- obuf
);
175 olen
+= stream
.next_out
- obuf
;
178 die(_("deflate error (%d)"), zret
);
180 if (zret
!= Z_STREAM_END
)
181 die(_("deflate error (%d)"), zret
);
185 git_deflate_end(&stream
);
190 * we are going to reuse the existing object data as is. make
191 * sure it is not corrupt.
193 static int check_pack_inflate(struct packed_git
*p
,
194 struct pack_window
**w_curs
,
197 unsigned long expect
)
200 unsigned char fakebuf
[4096], *in
;
203 memset(&stream
, 0, sizeof(stream
));
204 git_inflate_init(&stream
);
206 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
208 stream
.next_out
= fakebuf
;
209 stream
.avail_out
= sizeof(fakebuf
);
210 st
= git_inflate(&stream
, Z_FINISH
);
211 offset
+= stream
.next_in
- in
;
212 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
213 git_inflate_end(&stream
);
214 return (st
== Z_STREAM_END
&&
215 stream
.total_out
== expect
&&
216 stream
.total_in
== len
) ? 0 : -1;
219 static void copy_pack_data(struct sha1file
*f
,
220 struct packed_git
*p
,
221 struct pack_window
**w_curs
,
229 in
= use_pack(p
, w_curs
, offset
, &avail
);
231 avail
= (unsigned long)len
;
232 sha1write(f
, in
, avail
);
238 /* Return 0 if we will bust the pack-size limit */
239 static unsigned long write_no_reuse_object(struct sha1file
*f
, struct object_entry
*entry
,
240 unsigned long limit
, int usable_delta
)
242 unsigned long size
, datalen
;
243 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
244 dheader
[MAX_PACK_OBJECT_HEADER
];
246 enum object_type type
;
248 struct git_istream
*st
= NULL
;
251 if (entry
->type
== OBJ_BLOB
&&
252 entry
->size
> big_file_threshold
&&
253 (st
= open_istream(entry
->idx
.sha1
, &type
, &size
, NULL
)) != NULL
)
256 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
258 die(_("unable to read %s"), sha1_to_hex(entry
->idx
.sha1
));
261 * make sure no cached delta data remains from a
262 * previous attempt before a pack split occurred.
264 free(entry
->delta_data
);
265 entry
->delta_data
= NULL
;
266 entry
->z_delta_size
= 0;
267 } else if (entry
->delta_data
) {
268 size
= entry
->delta_size
;
269 buf
= entry
->delta_data
;
270 entry
->delta_data
= NULL
;
271 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
272 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
274 buf
= get_delta(entry
);
275 size
= entry
->delta_size
;
276 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
277 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
280 if (st
) /* large blob case, just assume we don't compress well */
282 else if (entry
->z_delta_size
)
283 datalen
= entry
->z_delta_size
;
285 datalen
= do_compress(&buf
, size
);
288 * The object header is a byte of 'type' followed by zero or
289 * more bytes of length.
291 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
294 if (type
== OBJ_OFS_DELTA
) {
296 * Deltas with relative base contain an additional
297 * encoding of the relative offset for the delta
298 * base from this object's position in the pack.
300 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
301 unsigned pos
= sizeof(dheader
) - 1;
302 dheader
[pos
] = ofs
& 127;
304 dheader
[--pos
] = 128 | (--ofs
& 127);
305 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
311 sha1write(f
, header
, hdrlen
);
312 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
313 hdrlen
+= sizeof(dheader
) - pos
;
314 } else if (type
== OBJ_REF_DELTA
) {
316 * Deltas with a base reference contain
317 * an additional 20 bytes for the base sha1.
319 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
325 sha1write(f
, header
, hdrlen
);
326 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
329 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
335 sha1write(f
, header
, hdrlen
);
338 datalen
= write_large_blob_data(st
, f
, entry
->idx
.sha1
);
341 sha1write(f
, buf
, datalen
);
345 return hdrlen
+ datalen
;
348 /* Return 0 if we will bust the pack-size limit */
349 static off_t
write_reuse_object(struct sha1file
*f
, struct object_entry
*entry
,
350 unsigned long limit
, int usable_delta
)
352 struct packed_git
*p
= entry
->in_pack
;
353 struct pack_window
*w_curs
= NULL
;
354 struct revindex_entry
*revidx
;
356 enum object_type type
= entry
->type
;
358 unsigned char header
[MAX_PACK_OBJECT_HEADER
],
359 dheader
[MAX_PACK_OBJECT_HEADER
];
363 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
364 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
365 hdrlen
= encode_in_pack_object_header(header
, sizeof(header
),
368 offset
= entry
->in_pack_offset
;
369 revidx
= find_pack_revindex(p
, offset
);
370 datalen
= revidx
[1].offset
- offset
;
371 if (!pack_to_stdout
&& p
->index_version
> 1 &&
372 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
)) {
373 error("bad packed object CRC for %s", sha1_to_hex(entry
->idx
.sha1
));
375 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
378 offset
+= entry
->in_pack_header_size
;
379 datalen
-= entry
->in_pack_header_size
;
381 if (!pack_to_stdout
&& p
->index_version
== 1 &&
382 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry
->size
)) {
383 error("corrupt packed object for %s", sha1_to_hex(entry
->idx
.sha1
));
385 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
388 if (type
== OBJ_OFS_DELTA
) {
389 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
390 unsigned pos
= sizeof(dheader
) - 1;
391 dheader
[pos
] = ofs
& 127;
393 dheader
[--pos
] = 128 | (--ofs
& 127);
394 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
398 sha1write(f
, header
, hdrlen
);
399 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
400 hdrlen
+= sizeof(dheader
) - pos
;
402 } else if (type
== OBJ_REF_DELTA
) {
403 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
407 sha1write(f
, header
, hdrlen
);
408 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
412 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
416 sha1write(f
, header
, hdrlen
);
418 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
421 return hdrlen
+ datalen
;
424 /* Return 0 if we will bust the pack-size limit */
425 static off_t
write_object(struct sha1file
*f
,
426 struct object_entry
*entry
,
431 int usable_delta
, to_reuse
;
436 /* apply size limit if limited packsize and not first object */
437 if (!pack_size_limit
|| !nr_written
)
439 else if (pack_size_limit
<= write_offset
)
441 * the earlier object did not fit the limit; avoid
442 * mistaking this with unlimited (i.e. limit = 0).
446 limit
= pack_size_limit
- write_offset
;
449 usable_delta
= 0; /* no delta */
450 else if (!pack_size_limit
)
451 usable_delta
= 1; /* unlimited packfile */
452 else if (entry
->delta
->idx
.offset
== (off_t
)-1)
453 usable_delta
= 0; /* base was written to another pack */
454 else if (entry
->delta
->idx
.offset
)
455 usable_delta
= 1; /* base already exists in this pack */
457 usable_delta
= 0; /* base could end up in another pack */
460 to_reuse
= 0; /* explicit */
461 else if (!entry
->in_pack
)
462 to_reuse
= 0; /* can't reuse what we don't have */
463 else if (entry
->type
== OBJ_REF_DELTA
|| entry
->type
== OBJ_OFS_DELTA
)
464 /* check_object() decided it for us ... */
465 to_reuse
= usable_delta
;
466 /* ... but pack split may override that */
467 else if (entry
->type
!= entry
->in_pack_type
)
468 to_reuse
= 0; /* pack has delta which is unusable */
469 else if (entry
->delta
)
470 to_reuse
= 0; /* we want to pack afresh */
472 to_reuse
= 1; /* we have it in-pack undeltified,
473 * and we do not need to deltify it.
477 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
479 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
487 entry
->idx
.crc32
= crc32_end(f
);
491 enum write_one_status
{
492 WRITE_ONE_SKIP
= -1, /* already written */
493 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
494 WRITE_ONE_WRITTEN
= 1, /* normal */
495 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
498 static enum write_one_status
write_one(struct sha1file
*f
,
499 struct object_entry
*e
,
506 * we set offset to 1 (which is an impossible value) to mark
507 * the fact that this object is involved in "write its base
508 * first before writing a deltified object" recursion.
510 recursing
= (e
->idx
.offset
== 1);
512 warning("recursive delta detected for object %s",
513 sha1_to_hex(e
->idx
.sha1
));
514 return WRITE_ONE_RECURSIVE
;
515 } else if (e
->idx
.offset
|| e
->preferred_base
) {
516 /* offset is non zero if object is written already. */
517 return WRITE_ONE_SKIP
;
520 /* if we are deltified, write out base object first. */
522 e
->idx
.offset
= 1; /* now recurse */
523 switch (write_one(f
, e
->delta
, offset
)) {
524 case WRITE_ONE_RECURSIVE
:
525 /* we cannot depend on this one */
530 case WRITE_ONE_BREAK
:
531 e
->idx
.offset
= recursing
;
532 return WRITE_ONE_BREAK
;
536 e
->idx
.offset
= *offset
;
537 size
= write_object(f
, e
, *offset
);
539 e
->idx
.offset
= recursing
;
540 return WRITE_ONE_BREAK
;
542 written_list
[nr_written
++] = &e
->idx
;
544 /* make sure off_t is sufficiently large not to wrap */
545 if (signed_add_overflows(*offset
, size
))
546 die("pack too large for current definition of off_t");
548 return WRITE_ONE_WRITTEN
;
551 static int mark_tagged(const char *path
, const struct object_id
*oid
, int flag
,
554 unsigned char peeled
[20];
555 struct object_entry
*entry
= packlist_find(&to_pack
, oid
->hash
, NULL
);
559 if (!peel_ref(path
, peeled
)) {
560 entry
= packlist_find(&to_pack
, peeled
, NULL
);
567 static inline void add_to_write_order(struct object_entry
**wo
,
569 struct object_entry
*e
)
577 static void add_descendants_to_write_order(struct object_entry
**wo
,
579 struct object_entry
*e
)
581 int add_to_order
= 1;
584 struct object_entry
*s
;
585 /* add this node... */
586 add_to_write_order(wo
, endp
, e
);
587 /* all its siblings... */
588 for (s
= e
->delta_sibling
; s
; s
= s
->delta_sibling
) {
589 add_to_write_order(wo
, endp
, s
);
592 /* drop down a level to add left subtree nodes if possible */
593 if (e
->delta_child
) {
598 /* our sibling might have some children, it is next */
599 if (e
->delta_sibling
) {
600 e
= e
->delta_sibling
;
603 /* go back to our parent node */
605 while (e
&& !e
->delta_sibling
) {
606 /* we're on the right side of a subtree, keep
607 * going up until we can go right again */
611 /* done- we hit our original root node */
614 /* pass it off to sibling at this level */
615 e
= e
->delta_sibling
;
620 static void add_family_to_write_order(struct object_entry
**wo
,
622 struct object_entry
*e
)
624 struct object_entry
*root
;
626 for (root
= e
; root
->delta
; root
= root
->delta
)
628 add_descendants_to_write_order(wo
, endp
, root
);
631 static struct object_entry
**compute_write_order(void)
633 unsigned int i
, wo_end
, last_untagged
;
635 struct object_entry
**wo
;
636 struct object_entry
*objects
= to_pack
.objects
;
638 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
639 objects
[i
].tagged
= 0;
640 objects
[i
].filled
= 0;
641 objects
[i
].delta_child
= NULL
;
642 objects
[i
].delta_sibling
= NULL
;
646 * Fully connect delta_child/delta_sibling network.
647 * Make sure delta_sibling is sorted in the original
650 for (i
= to_pack
.nr_objects
; i
> 0;) {
651 struct object_entry
*e
= &objects
[--i
];
654 /* Mark me as the first child */
655 e
->delta_sibling
= e
->delta
->delta_child
;
656 e
->delta
->delta_child
= e
;
660 * Mark objects that are at the tip of tags.
662 for_each_tag_ref(mark_tagged
, NULL
);
665 * Give the objects in the original recency order until
666 * we see a tagged tip.
668 ALLOC_ARRAY(wo
, to_pack
.nr_objects
);
669 for (i
= wo_end
= 0; i
< to_pack
.nr_objects
; i
++) {
670 if (objects
[i
].tagged
)
672 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
677 * Then fill all the tagged tips.
679 for (; i
< to_pack
.nr_objects
; i
++) {
680 if (objects
[i
].tagged
)
681 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
685 * And then all remaining commits and tags.
687 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
688 if (objects
[i
].type
!= OBJ_COMMIT
&&
689 objects
[i
].type
!= OBJ_TAG
)
691 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
695 * And then all the trees.
697 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
698 if (objects
[i
].type
!= OBJ_TREE
)
700 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
704 * Finally all the rest in really tight order
706 for (i
= last_untagged
; i
< to_pack
.nr_objects
; i
++) {
707 if (!objects
[i
].filled
)
708 add_family_to_write_order(wo
, &wo_end
, &objects
[i
]);
711 if (wo_end
!= to_pack
.nr_objects
)
712 die("ordered %u objects, expected %"PRIu32
, wo_end
, to_pack
.nr_objects
);
717 static off_t
write_reused_pack(struct sha1file
*f
)
719 unsigned char buffer
[8192];
720 off_t to_write
, total
;
723 if (!is_pack_valid(reuse_packfile
))
724 die("packfile is invalid: %s", reuse_packfile
->pack_name
);
726 fd
= git_open(reuse_packfile
->pack_name
);
728 die_errno("unable to open packfile for reuse: %s",
729 reuse_packfile
->pack_name
);
731 if (lseek(fd
, sizeof(struct pack_header
), SEEK_SET
) == -1)
732 die_errno("unable to seek in reused packfile");
734 if (reuse_packfile_offset
< 0)
735 reuse_packfile_offset
= reuse_packfile
->pack_size
- 20;
737 total
= to_write
= reuse_packfile_offset
- sizeof(struct pack_header
);
740 int read_pack
= xread(fd
, buffer
, sizeof(buffer
));
743 die_errno("unable to read from reused packfile");
745 if (read_pack
> to_write
)
746 read_pack
= to_write
;
748 sha1write(f
, buffer
, read_pack
);
749 to_write
-= read_pack
;
752 * We don't know the actual number of objects written,
753 * only how many bytes written, how many bytes total, and
754 * how many objects total. So we can fake it by pretending all
755 * objects we are writing are the same size. This gives us a
756 * smooth progress meter, and at the end it matches the true
759 written
= reuse_packfile_objects
*
760 (((double)(total
- to_write
)) / total
);
761 display_progress(progress_state
, written
);
765 written
= reuse_packfile_objects
;
766 display_progress(progress_state
, written
);
767 return reuse_packfile_offset
- sizeof(struct pack_header
);
770 static const char no_split_warning
[] = N_(
771 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
774 static void write_pack_file(void)
779 uint32_t nr_remaining
= nr_result
;
780 time_t last_mtime
= 0;
781 struct object_entry
**write_order
;
783 if (progress
> pack_to_stdout
)
784 progress_state
= start_progress(_("Writing objects"), nr_result
);
785 ALLOC_ARRAY(written_list
, to_pack
.nr_objects
);
786 write_order
= compute_write_order();
789 unsigned char sha1
[20];
790 char *pack_tmp_name
= NULL
;
793 f
= sha1fd_throughput(1, "<stdout>", progress_state
);
795 f
= create_tmp_packfile(&pack_tmp_name
);
797 offset
= write_pack_header(f
, nr_remaining
);
799 if (reuse_packfile
) {
801 assert(pack_to_stdout
);
803 packfile_size
= write_reused_pack(f
);
804 offset
+= packfile_size
;
808 for (; i
< to_pack
.nr_objects
; i
++) {
809 struct object_entry
*e
= write_order
[i
];
810 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
812 display_progress(progress_state
, written
);
816 * Did we write the wrong # entries in the header?
817 * If so, rewrite it like in fast-import
819 if (pack_to_stdout
) {
820 sha1close(f
, sha1
, CSUM_CLOSE
);
821 } else if (nr_written
== nr_remaining
) {
822 sha1close(f
, sha1
, CSUM_FSYNC
);
824 int fd
= sha1close(f
, sha1
, 0);
825 fixup_pack_header_footer(fd
, sha1
, pack_tmp_name
,
826 nr_written
, sha1
, offset
);
828 if (write_bitmap_index
) {
829 warning(_(no_split_warning
));
830 write_bitmap_index
= 0;
834 if (!pack_to_stdout
) {
836 struct strbuf tmpname
= STRBUF_INIT
;
839 * Packs are runtime accessed in their mtime
840 * order since newer packs are more likely to contain
841 * younger objects. So if we are creating multiple
842 * packs then we should modify the mtime of later ones
843 * to preserve this property.
845 if (stat(pack_tmp_name
, &st
) < 0) {
846 warning_errno("failed to stat %s", pack_tmp_name
);
847 } else if (!last_mtime
) {
848 last_mtime
= st
.st_mtime
;
851 utb
.actime
= st
.st_atime
;
852 utb
.modtime
= --last_mtime
;
853 if (utime(pack_tmp_name
, &utb
) < 0)
854 warning_errno("failed utime() on %s", pack_tmp_name
);
857 strbuf_addf(&tmpname
, "%s-", base_name
);
859 if (write_bitmap_index
) {
860 bitmap_writer_set_checksum(sha1
);
861 bitmap_writer_build_type_index(written_list
, nr_written
);
864 finish_tmp_packfile(&tmpname
, pack_tmp_name
,
865 written_list
, nr_written
,
866 &pack_idx_opts
, sha1
);
868 if (write_bitmap_index
) {
869 strbuf_addf(&tmpname
, "%s.bitmap", sha1_to_hex(sha1
));
871 stop_progress(&progress_state
);
873 bitmap_writer_show_progress(progress
);
874 bitmap_writer_reuse_bitmaps(&to_pack
);
875 bitmap_writer_select_commits(indexed_commits
, indexed_commits_nr
, -1);
876 bitmap_writer_build(&to_pack
);
877 bitmap_writer_finish(written_list
, nr_written
,
878 tmpname
.buf
, write_bitmap_options
);
879 write_bitmap_index
= 0;
882 strbuf_release(&tmpname
);
884 puts(sha1_to_hex(sha1
));
887 /* mark written objects as written to previous pack */
888 for (j
= 0; j
< nr_written
; j
++) {
889 written_list
[j
]->offset
= (off_t
)-1;
891 nr_remaining
-= nr_written
;
892 } while (nr_remaining
&& i
< to_pack
.nr_objects
);
896 stop_progress(&progress_state
);
897 if (written
!= nr_result
)
898 die("wrote %"PRIu32
" objects while expecting %"PRIu32
,
902 static int no_try_delta(const char *path
)
904 static struct attr_check
*check
;
907 check
= attr_check_initl("delta", NULL
);
908 if (git_check_attr(path
, check
))
910 if (ATTR_FALSE(check
->items
[0].value
))
916 * When adding an object, check whether we have already added it
917 * to our packing list. If so, we can skip. However, if we are
918 * being asked to excludei t, but the previous mention was to include
919 * it, make sure to adjust its flags and tweak our numbers accordingly.
921 * As an optimization, we pass out the index position where we would have
922 * found the item, since that saves us from having to look it up again a
923 * few lines later when we want to add the new entry.
925 static int have_duplicate_entry(const unsigned char *sha1
,
929 struct object_entry
*entry
;
931 entry
= packlist_find(&to_pack
, sha1
, index_pos
);
936 if (!entry
->preferred_base
)
938 entry
->preferred_base
= 1;
944 static int want_found_object(int exclude
, struct packed_git
*p
)
952 * When asked to do --local (do not include an object that appears in a
953 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
954 * an object that appears in a pack marked with .keep), finding a pack
955 * that matches the criteria is sufficient for us to decide to omit it.
956 * However, even if this pack does not satisfy the criteria, we need to
957 * make sure no copy of this object appears in _any_ pack that makes us
958 * to omit the object, so we need to check all the packs.
960 * We can however first check whether these options can possible matter;
961 * if they do not matter we know we want the object in generated pack.
962 * Otherwise, we signal "-1" at the end to tell the caller that we do
963 * not know either way, and it needs to check more packs.
965 if (!ignore_packed_keep
&&
966 (!local
|| !have_non_local_packs
))
969 if (local
&& !p
->pack_local
)
971 if (ignore_packed_keep
&& p
->pack_local
&& p
->pack_keep
)
974 /* we don't know yet; keep looking for more packs */
979 * Check whether we want the object in the pack (e.g., we do not want
980 * objects found in non-local stores if the "--local" option was used).
982 * If the caller already knows an existing pack it wants to take the object
983 * from, that is passed in *found_pack and *found_offset; otherwise this
984 * function finds if there is any pack that has the object and returns the pack
985 * and its offset in these variables.
987 static int want_object_in_pack(const unsigned char *sha1
,
989 struct packed_git
**found_pack
,
992 struct mru_entry
*entry
;
995 if (!exclude
&& local
&& has_loose_object_nonlocal(sha1
))
999 * If we already know the pack object lives in, start checks from that
1000 * pack - in the usual case when neither --local was given nor .keep files
1001 * are present we will determine the answer right now.
1004 want
= want_found_object(exclude
, *found_pack
);
1009 for (entry
= packed_git_mru
->head
; entry
; entry
= entry
->next
) {
1010 struct packed_git
*p
= entry
->item
;
1013 if (p
== *found_pack
)
1014 offset
= *found_offset
;
1016 offset
= find_pack_entry_one(sha1
, p
);
1020 if (!is_pack_valid(p
))
1022 *found_offset
= offset
;
1025 want
= want_found_object(exclude
, p
);
1026 if (!exclude
&& want
> 0)
1027 mru_mark(packed_git_mru
, entry
);
1036 static void create_object_entry(const unsigned char *sha1
,
1037 enum object_type type
,
1042 struct packed_git
*found_pack
,
1045 struct object_entry
*entry
;
1047 entry
= packlist_alloc(&to_pack
, sha1
, index_pos
);
1052 entry
->preferred_base
= 1;
1056 entry
->in_pack
= found_pack
;
1057 entry
->in_pack_offset
= found_offset
;
1060 entry
->no_try_delta
= no_try_delta
;
1063 static const char no_closure_warning
[] = N_(
1064 "disabling bitmap writing, as some objects are not being packed"
1067 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
1068 const char *name
, int exclude
)
1070 struct packed_git
*found_pack
= NULL
;
1071 off_t found_offset
= 0;
1074 if (have_duplicate_entry(sha1
, exclude
, &index_pos
))
1077 if (!want_object_in_pack(sha1
, exclude
, &found_pack
, &found_offset
)) {
1078 /* The pack is missing an object, so it will not have closure */
1079 if (write_bitmap_index
) {
1080 warning(_(no_closure_warning
));
1081 write_bitmap_index
= 0;
1086 create_object_entry(sha1
, type
, pack_name_hash(name
),
1087 exclude
, name
&& no_try_delta(name
),
1088 index_pos
, found_pack
, found_offset
);
1090 display_progress(progress_state
, nr_result
);
1094 static int add_object_entry_from_bitmap(const unsigned char *sha1
,
1095 enum object_type type
,
1096 int flags
, uint32_t name_hash
,
1097 struct packed_git
*pack
, off_t offset
)
1101 if (have_duplicate_entry(sha1
, 0, &index_pos
))
1104 if (!want_object_in_pack(sha1
, 0, &pack
, &offset
))
1107 create_object_entry(sha1
, type
, name_hash
, 0, 0, index_pos
, pack
, offset
);
1109 display_progress(progress_state
, nr_result
);
1113 struct pbase_tree_cache
{
1114 unsigned char sha1
[20];
1118 unsigned long tree_size
;
1121 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
1122 static int pbase_tree_cache_ix(const unsigned char *sha1
)
1124 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
1126 static int pbase_tree_cache_ix_incr(int ix
)
1128 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
1131 static struct pbase_tree
{
1132 struct pbase_tree
*next
;
1133 /* This is a phony "cache" entry; we are not
1134 * going to evict it or find it through _get()
1135 * mechanism -- this is for the toplevel node that
1136 * would almost always change with any commit.
1138 struct pbase_tree_cache pcache
;
1141 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
1143 struct pbase_tree_cache
*ent
, *nent
;
1146 enum object_type type
;
1148 int my_ix
= pbase_tree_cache_ix(sha1
);
1149 int available_ix
= -1;
1151 /* pbase-tree-cache acts as a limited hashtable.
1152 * your object will be found at your index or within a few
1153 * slots after that slot if it is cached.
1155 for (neigh
= 0; neigh
< 8; neigh
++) {
1156 ent
= pbase_tree_cache
[my_ix
];
1157 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
1161 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
1162 ((0 <= available_ix
) &&
1163 (!ent
&& pbase_tree_cache
[available_ix
])))
1164 available_ix
= my_ix
;
1167 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
1170 /* Did not find one. Either we got a bogus request or
1171 * we need to read and perhaps cache.
1173 data
= read_sha1_file(sha1
, &type
, &size
);
1176 if (type
!= OBJ_TREE
) {
1181 /* We need to either cache or return a throwaway copy */
1183 if (available_ix
< 0)
1186 ent
= pbase_tree_cache
[available_ix
];
1187 my_ix
= available_ix
;
1191 nent
= xmalloc(sizeof(*nent
));
1192 nent
->temporary
= (available_ix
< 0);
1195 /* evict and reuse */
1196 free(ent
->tree_data
);
1199 hashcpy(nent
->sha1
, sha1
);
1200 nent
->tree_data
= data
;
1201 nent
->tree_size
= size
;
1203 if (!nent
->temporary
)
1204 pbase_tree_cache
[my_ix
] = nent
;
1208 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1210 if (!cache
->temporary
) {
1214 free(cache
->tree_data
);
1218 static int name_cmp_len(const char *name
)
1221 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
1226 static void add_pbase_object(struct tree_desc
*tree
,
1229 const char *fullname
)
1231 struct name_entry entry
;
1234 while (tree_entry(tree
,&entry
)) {
1235 if (S_ISGITLINK(entry
.mode
))
1237 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1238 memcmp(name
, entry
.path
, cmplen
);
1243 if (name
[cmplen
] != '/') {
1244 add_object_entry(entry
.oid
->hash
,
1245 object_type(entry
.mode
),
1249 if (S_ISDIR(entry
.mode
)) {
1250 struct tree_desc sub
;
1251 struct pbase_tree_cache
*tree
;
1252 const char *down
= name
+cmplen
+1;
1253 int downlen
= name_cmp_len(down
);
1255 tree
= pbase_tree_get(entry
.oid
->hash
);
1258 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1260 add_pbase_object(&sub
, down
, downlen
, fullname
);
1261 pbase_tree_put(tree
);
1266 static unsigned *done_pbase_paths
;
1267 static int done_pbase_paths_num
;
1268 static int done_pbase_paths_alloc
;
1269 static int done_pbase_path_pos(unsigned hash
)
1272 int hi
= done_pbase_paths_num
;
1274 int mi
= (hi
+ lo
) / 2;
1275 if (done_pbase_paths
[mi
] == hash
)
1277 if (done_pbase_paths
[mi
] < hash
)
1285 static int check_pbase_path(unsigned hash
)
1287 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
1291 ALLOC_GROW(done_pbase_paths
,
1292 done_pbase_paths_num
+ 1,
1293 done_pbase_paths_alloc
);
1294 done_pbase_paths_num
++;
1295 if (pos
< done_pbase_paths_num
)
1296 memmove(done_pbase_paths
+ pos
+ 1,
1297 done_pbase_paths
+ pos
,
1298 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
1299 done_pbase_paths
[pos
] = hash
;
1303 static void add_preferred_base_object(const char *name
)
1305 struct pbase_tree
*it
;
1307 unsigned hash
= pack_name_hash(name
);
1309 if (!num_preferred_base
|| check_pbase_path(hash
))
1312 cmplen
= name_cmp_len(name
);
1313 for (it
= pbase_tree
; it
; it
= it
->next
) {
1315 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
1318 struct tree_desc tree
;
1319 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1320 add_pbase_object(&tree
, name
, cmplen
, name
);
1325 static void add_preferred_base(unsigned char *sha1
)
1327 struct pbase_tree
*it
;
1330 unsigned char tree_sha1
[20];
1332 if (window
<= num_preferred_base
++)
1335 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
1339 for (it
= pbase_tree
; it
; it
= it
->next
) {
1340 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
1346 it
= xcalloc(1, sizeof(*it
));
1347 it
->next
= pbase_tree
;
1350 hashcpy(it
->pcache
.sha1
, tree_sha1
);
1351 it
->pcache
.tree_data
= data
;
1352 it
->pcache
.tree_size
= size
;
1355 static void cleanup_preferred_base(void)
1357 struct pbase_tree
*it
;
1363 struct pbase_tree
*this = it
;
1365 free(this->pcache
.tree_data
);
1369 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1370 if (!pbase_tree_cache
[i
])
1372 free(pbase_tree_cache
[i
]->tree_data
);
1373 free(pbase_tree_cache
[i
]);
1374 pbase_tree_cache
[i
] = NULL
;
1377 free(done_pbase_paths
);
1378 done_pbase_paths
= NULL
;
1379 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1382 static void check_object(struct object_entry
*entry
)
1384 if (entry
->in_pack
) {
1385 struct packed_git
*p
= entry
->in_pack
;
1386 struct pack_window
*w_curs
= NULL
;
1387 const unsigned char *base_ref
= NULL
;
1388 struct object_entry
*base_entry
;
1389 unsigned long used
, used_0
;
1390 unsigned long avail
;
1392 unsigned char *buf
, c
;
1394 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1397 * We want in_pack_type even if we do not reuse delta
1398 * since non-delta representations could still be reused.
1400 used
= unpack_object_header_buffer(buf
, avail
,
1401 &entry
->in_pack_type
,
1407 * Determine if this is a delta and if so whether we can
1408 * reuse it or not. Otherwise let's find out as cheaply as
1409 * possible what the actual type and size for this object is.
1411 switch (entry
->in_pack_type
) {
1413 /* Not a delta hence we've already got all we need. */
1414 entry
->type
= entry
->in_pack_type
;
1415 entry
->in_pack_header_size
= used
;
1416 if (entry
->type
< OBJ_COMMIT
|| entry
->type
> OBJ_BLOB
)
1418 unuse_pack(&w_curs
);
1421 if (reuse_delta
&& !entry
->preferred_base
)
1422 base_ref
= use_pack(p
, &w_curs
,
1423 entry
->in_pack_offset
+ used
, NULL
);
1424 entry
->in_pack_header_size
= used
+ 20;
1427 buf
= use_pack(p
, &w_curs
,
1428 entry
->in_pack_offset
+ used
, NULL
);
1434 if (!ofs
|| MSB(ofs
, 7)) {
1435 error("delta base offset overflow in pack for %s",
1436 sha1_to_hex(entry
->idx
.sha1
));
1440 ofs
= (ofs
<< 7) + (c
& 127);
1442 ofs
= entry
->in_pack_offset
- ofs
;
1443 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1444 error("delta base offset out of bound for %s",
1445 sha1_to_hex(entry
->idx
.sha1
));
1448 if (reuse_delta
&& !entry
->preferred_base
) {
1449 struct revindex_entry
*revidx
;
1450 revidx
= find_pack_revindex(p
, ofs
);
1453 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1455 entry
->in_pack_header_size
= used
+ used_0
;
1459 if (base_ref
&& (base_entry
= packlist_find(&to_pack
, base_ref
, NULL
))) {
1461 * If base_ref was set above that means we wish to
1462 * reuse delta data, and we even found that base
1463 * in the list of objects we want to pack. Goodie!
1465 * Depth value does not matter - find_deltas() will
1466 * never consider reused delta as the base object to
1467 * deltify other objects against, in order to avoid
1470 entry
->type
= entry
->in_pack_type
;
1471 entry
->delta
= base_entry
;
1472 entry
->delta_size
= entry
->size
;
1473 entry
->delta_sibling
= base_entry
->delta_child
;
1474 base_entry
->delta_child
= entry
;
1475 unuse_pack(&w_curs
);
1481 * This must be a delta and we already know what the
1482 * final object type is. Let's extract the actual
1483 * object size from the delta header.
1485 entry
->size
= get_size_from_delta(p
, &w_curs
,
1486 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1487 if (entry
->size
== 0)
1489 unuse_pack(&w_curs
);
1494 * No choice but to fall back to the recursive delta walk
1495 * with sha1_object_info() to find about the object type
1499 unuse_pack(&w_curs
);
1502 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1504 * The error condition is checked in prepare_pack(). This is
1505 * to permit a missing preferred base object to be ignored
1506 * as a preferred base. Doing so can result in a larger
1507 * pack file, but the transfer will still take place.
1511 static int pack_offset_sort(const void *_a
, const void *_b
)
1513 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1514 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1516 /* avoid filesystem trashing with loose objects */
1517 if (!a
->in_pack
&& !b
->in_pack
)
1518 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1520 if (a
->in_pack
< b
->in_pack
)
1522 if (a
->in_pack
> b
->in_pack
)
1524 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1525 (a
->in_pack_offset
> b
->in_pack_offset
);
1529 * Drop an on-disk delta we were planning to reuse. Naively, this would
1530 * just involve blanking out the "delta" field, but we have to deal
1531 * with some extra book-keeping:
1533 * 1. Removing ourselves from the delta_sibling linked list.
1535 * 2. Updating our size/type to the non-delta representation. These were
1536 * either not recorded initially (size) or overwritten with the delta type
1537 * (type) when check_object() decided to reuse the delta.
1539 * 3. Resetting our delta depth, as we are now a base object.
1541 static void drop_reused_delta(struct object_entry
*entry
)
1543 struct object_entry
**p
= &entry
->delta
->delta_child
;
1544 struct object_info oi
= OBJECT_INFO_INIT
;
1548 *p
= (*p
)->delta_sibling
;
1550 p
= &(*p
)->delta_sibling
;
1552 entry
->delta
= NULL
;
1555 oi
.sizep
= &entry
->size
;
1556 oi
.typep
= &entry
->type
;
1557 if (packed_object_info(entry
->in_pack
, entry
->in_pack_offset
, &oi
) < 0) {
1559 * We failed to get the info from this pack for some reason;
1560 * fall back to sha1_object_info, which may find another copy.
1561 * And if that fails, the error will be recorded in entry->type
1562 * and dealt with in prepare_pack().
1564 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1569 * Follow the chain of deltas from this entry onward, throwing away any links
1570 * that cause us to hit a cycle (as determined by the DFS state flags in
1573 * We also detect too-long reused chains that would violate our --depth
1576 static void break_delta_chains(struct object_entry
*entry
)
1579 * The actual depth of each object we will write is stored as an int,
1580 * as it cannot exceed our int "depth" limit. But before we break
1581 * changes based no that limit, we may potentially go as deep as the
1582 * number of objects, which is elsewhere bounded to a uint32_t.
1584 uint32_t total_depth
;
1585 struct object_entry
*cur
, *next
;
1587 for (cur
= entry
, total_depth
= 0;
1589 cur
= cur
->delta
, total_depth
++) {
1590 if (cur
->dfs_state
== DFS_DONE
) {
1592 * We've already seen this object and know it isn't
1593 * part of a cycle. We do need to append its depth
1596 total_depth
+= cur
->depth
;
1601 * We break cycles before looping, so an ACTIVE state (or any
1602 * other cruft which made its way into the state variable)
1605 if (cur
->dfs_state
!= DFS_NONE
)
1606 die("BUG: confusing delta dfs state in first pass: %d",
1610 * Now we know this is the first time we've seen the object. If
1611 * it's not a delta, we're done traversing, but we'll mark it
1612 * done to save time on future traversals.
1615 cur
->dfs_state
= DFS_DONE
;
1620 * Mark ourselves as active and see if the next step causes
1621 * us to cycle to another active object. It's important to do
1622 * this _before_ we loop, because it impacts where we make the
1623 * cut, and thus how our total_depth counter works.
1624 * E.g., We may see a partial loop like:
1626 * A -> B -> C -> D -> B
1628 * Cutting B->C breaks the cycle. But now the depth of A is
1629 * only 1, and our total_depth counter is at 3. The size of the
1630 * error is always one less than the size of the cycle we
1631 * broke. Commits C and D were "lost" from A's chain.
1633 * If we instead cut D->B, then the depth of A is correct at 3.
1634 * We keep all commits in the chain that we examined.
1636 cur
->dfs_state
= DFS_ACTIVE
;
1637 if (cur
->delta
->dfs_state
== DFS_ACTIVE
) {
1638 drop_reused_delta(cur
);
1639 cur
->dfs_state
= DFS_DONE
;
1645 * And now that we've gone all the way to the bottom of the chain, we
1646 * need to clear the active flags and set the depth fields as
1647 * appropriate. Unlike the loop above, which can quit when it drops a
1648 * delta, we need to keep going to look for more depth cuts. So we need
1649 * an extra "next" pointer to keep going after we reset cur->delta.
1651 for (cur
= entry
; cur
; cur
= next
) {
1655 * We should have a chain of zero or more ACTIVE states down to
1656 * a final DONE. We can quit after the DONE, because either it
1657 * has no bases, or we've already handled them in a previous
1660 if (cur
->dfs_state
== DFS_DONE
)
1662 else if (cur
->dfs_state
!= DFS_ACTIVE
)
1663 die("BUG: confusing delta dfs state in second pass: %d",
1667 * If the total_depth is more than depth, then we need to snip
1668 * the chain into two or more smaller chains that don't exceed
1669 * the maximum depth. Most of the resulting chains will contain
1670 * (depth + 1) entries (i.e., depth deltas plus one base), and
1671 * the last chain (i.e., the one containing entry) will contain
1672 * whatever entries are left over, namely
1673 * (total_depth % (depth + 1)) of them.
1675 * Since we are iterating towards decreasing depth, we need to
1676 * decrement total_depth as we go, and we need to write to the
1677 * entry what its final depth will be after all of the
1678 * snipping. Since we're snipping into chains of length (depth
1679 * + 1) entries, the final depth of an entry will be its
1680 * original depth modulo (depth + 1). Any time we encounter an
1681 * entry whose final depth is supposed to be zero, we snip it
1682 * from its delta base, thereby making it so.
1684 cur
->depth
= (total_depth
--) % (depth
+ 1);
1686 drop_reused_delta(cur
);
1688 cur
->dfs_state
= DFS_DONE
;
1692 static void get_object_details(void)
1695 struct object_entry
**sorted_by_offset
;
1697 sorted_by_offset
= xcalloc(to_pack
.nr_objects
, sizeof(struct object_entry
*));
1698 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1699 sorted_by_offset
[i
] = to_pack
.objects
+ i
;
1700 QSORT(sorted_by_offset
, to_pack
.nr_objects
, pack_offset_sort
);
1702 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
1703 struct object_entry
*entry
= sorted_by_offset
[i
];
1704 check_object(entry
);
1705 if (big_file_threshold
< entry
->size
)
1706 entry
->no_try_delta
= 1;
1710 * This must happen in a second pass, since we rely on the delta
1711 * information for the whole list being completed.
1713 for (i
= 0; i
< to_pack
.nr_objects
; i
++)
1714 break_delta_chains(&to_pack
.objects
[i
]);
1716 free(sorted_by_offset
);
1720 * We search for deltas in a list sorted by type, by filename hash, and then
1721 * by size, so that we see progressively smaller and smaller files.
1722 * That's because we prefer deltas to be from the bigger file
1723 * to the smaller -- deletes are potentially cheaper, but perhaps
1724 * more importantly, the bigger file is likely the more recent
1725 * one. The deepest deltas are therefore the oldest objects which are
1726 * less susceptible to be accessed often.
1728 static int type_size_sort(const void *_a
, const void *_b
)
1730 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1731 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1733 if (a
->type
> b
->type
)
1735 if (a
->type
< b
->type
)
1737 if (a
->hash
> b
->hash
)
1739 if (a
->hash
< b
->hash
)
1741 if (a
->preferred_base
> b
->preferred_base
)
1743 if (a
->preferred_base
< b
->preferred_base
)
1745 if (a
->size
> b
->size
)
1747 if (a
->size
< b
->size
)
1749 return a
< b
? -1 : (a
> b
); /* newest first */
1753 struct object_entry
*entry
;
1755 struct delta_index
*index
;
1759 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1760 unsigned long delta_size
)
1762 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1765 if (delta_size
< cache_max_small_delta_size
)
1768 /* cache delta, if objects are large enough compared to delta size */
1769 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1777 static pthread_mutex_t read_mutex
;
1778 #define read_lock() pthread_mutex_lock(&read_mutex)
1779 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1781 static pthread_mutex_t cache_mutex
;
1782 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1783 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1785 static pthread_mutex_t progress_mutex
;
1786 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1787 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1791 #define read_lock() (void)0
1792 #define read_unlock() (void)0
1793 #define cache_lock() (void)0
1794 #define cache_unlock() (void)0
1795 #define progress_lock() (void)0
1796 #define progress_unlock() (void)0
1800 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1801 unsigned max_depth
, unsigned long *mem_usage
)
1803 struct object_entry
*trg_entry
= trg
->entry
;
1804 struct object_entry
*src_entry
= src
->entry
;
1805 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1807 enum object_type type
;
1810 /* Don't bother doing diffs between different types */
1811 if (trg_entry
->type
!= src_entry
->type
)
1815 * We do not bother to try a delta that we discarded on an
1816 * earlier try, but only when reusing delta data. Note that
1817 * src_entry that is marked as the preferred_base should always
1818 * be considered, as even if we produce a suboptimal delta against
1819 * it, we will still save the transfer cost, as we already know
1820 * the other side has it and we won't send src_entry at all.
1822 if (reuse_delta
&& trg_entry
->in_pack
&&
1823 trg_entry
->in_pack
== src_entry
->in_pack
&&
1824 !src_entry
->preferred_base
&&
1825 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1826 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1829 /* Let's not bust the allowed depth. */
1830 if (src
->depth
>= max_depth
)
1833 /* Now some size filtering heuristics. */
1834 trg_size
= trg_entry
->size
;
1835 if (!trg_entry
->delta
) {
1836 max_size
= trg_size
/2 - 20;
1839 max_size
= trg_entry
->delta_size
;
1840 ref_depth
= trg
->depth
;
1842 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
1843 (max_depth
- ref_depth
+ 1);
1846 src_size
= src_entry
->size
;
1847 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1848 if (sizediff
>= max_size
)
1850 if (trg_size
< src_size
/ 32)
1853 /* Load data if not already done */
1856 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1859 die("object %s cannot be read",
1860 sha1_to_hex(trg_entry
->idx
.sha1
));
1862 die("object %s inconsistent object length (%lu vs %lu)",
1863 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1868 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1871 if (src_entry
->preferred_base
) {
1872 static int warned
= 0;
1874 warning("object %s cannot be read",
1875 sha1_to_hex(src_entry
->idx
.sha1
));
1877 * Those objects are not included in the
1878 * resulting pack. Be resilient and ignore
1879 * them if they can't be read, in case the
1880 * pack could be created nevertheless.
1884 die("object %s cannot be read",
1885 sha1_to_hex(src_entry
->idx
.sha1
));
1888 die("object %s inconsistent object length (%lu vs %lu)",
1889 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1893 src
->index
= create_delta_index(src
->data
, src_size
);
1895 static int warned
= 0;
1897 warning("suboptimal pack - out of memory");
1900 *mem_usage
+= sizeof_delta_index(src
->index
);
1903 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1907 if (trg_entry
->delta
) {
1908 /* Prefer only shallower same-sized deltas. */
1909 if (delta_size
== trg_entry
->delta_size
&&
1910 src
->depth
+ 1 >= trg
->depth
) {
1917 * Handle memory allocation outside of the cache
1918 * accounting lock. Compiler will optimize the strangeness
1919 * away when NO_PTHREADS is defined.
1921 free(trg_entry
->delta_data
);
1923 if (trg_entry
->delta_data
) {
1924 delta_cache_size
-= trg_entry
->delta_size
;
1925 trg_entry
->delta_data
= NULL
;
1927 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1928 delta_cache_size
+= delta_size
;
1930 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1936 trg_entry
->delta
= src_entry
;
1937 trg_entry
->delta_size
= delta_size
;
1938 trg
->depth
= src
->depth
+ 1;
1943 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1945 struct object_entry
*child
= me
->delta_child
;
1948 unsigned int c
= check_delta_limit(child
, n
+ 1);
1951 child
= child
->delta_sibling
;
1956 static unsigned long free_unpacked(struct unpacked
*n
)
1958 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1959 free_delta_index(n
->index
);
1962 freed_mem
+= n
->entry
->size
;
1971 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1972 int window
, int depth
, unsigned *processed
)
1974 uint32_t i
, idx
= 0, count
= 0;
1975 struct unpacked
*array
;
1976 unsigned long mem_usage
= 0;
1978 array
= xcalloc(window
, sizeof(struct unpacked
));
1981 struct object_entry
*entry
;
1982 struct unpacked
*n
= array
+ idx
;
1983 int j
, max_depth
, best_base
= -1;
1992 if (!entry
->preferred_base
) {
1994 display_progress(progress_state
, *processed
);
1998 mem_usage
-= free_unpacked(n
);
2001 while (window_memory_limit
&&
2002 mem_usage
> window_memory_limit
&&
2004 uint32_t tail
= (idx
+ window
- count
) % window
;
2005 mem_usage
-= free_unpacked(array
+ tail
);
2009 /* We do not compute delta to *create* objects we are not
2012 if (entry
->preferred_base
)
2016 * If the current object is at pack edge, take the depth the
2017 * objects that depend on the current object into account
2018 * otherwise they would become too deep.
2021 if (entry
->delta_child
) {
2022 max_depth
-= check_delta_limit(entry
, 0);
2030 uint32_t other_idx
= idx
+ j
;
2032 if (other_idx
>= window
)
2033 other_idx
-= window
;
2034 m
= array
+ other_idx
;
2037 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
2041 best_base
= other_idx
;
2045 * If we decided to cache the delta data, then it is best
2046 * to compress it right away. First because we have to do
2047 * it anyway, and doing it here while we're threaded will
2048 * save a lot of time in the non threaded write phase,
2049 * as well as allow for caching more deltas within
2050 * the same cache size limit.
2052 * But only if not writing to stdout, since in that case
2053 * the network is most likely throttling writes anyway,
2054 * and therefore it is best to go to the write phase ASAP
2055 * instead, as we can afford spending more time compressing
2056 * between writes at that moment.
2058 if (entry
->delta_data
&& !pack_to_stdout
) {
2059 entry
->z_delta_size
= do_compress(&entry
->delta_data
,
2062 delta_cache_size
-= entry
->delta_size
;
2063 delta_cache_size
+= entry
->z_delta_size
;
2067 /* if we made n a delta, and if n is already at max
2068 * depth, leaving it in the window is pointless. we
2069 * should evict it first.
2071 if (entry
->delta
&& max_depth
<= n
->depth
)
2075 * Move the best delta base up in the window, after the
2076 * currently deltified object, to keep it longer. It will
2077 * be the first base object to be attempted next.
2080 struct unpacked swap
= array
[best_base
];
2081 int dist
= (window
+ idx
- best_base
) % window
;
2082 int dst
= best_base
;
2084 int src
= (dst
+ 1) % window
;
2085 array
[dst
] = array
[src
];
2093 if (count
+ 1 < window
)
2099 for (i
= 0; i
< window
; ++i
) {
2100 free_delta_index(array
[i
].index
);
2101 free(array
[i
].data
);
2108 static void try_to_free_from_threads(size_t size
)
2111 release_pack_memory(size
);
2115 static try_to_free_t old_try_to_free_routine
;
2118 * The main thread waits on the condition that (at least) one of the workers
2119 * has stopped working (which is indicated in the .working member of
2120 * struct thread_params).
2121 * When a work thread has completed its work, it sets .working to 0 and
2122 * signals the main thread and waits on the condition that .data_ready
2126 struct thread_params
{
2128 struct object_entry
**list
;
2135 pthread_mutex_t mutex
;
2136 pthread_cond_t cond
;
2137 unsigned *processed
;
2140 static pthread_cond_t progress_cond
;
2143 * Mutex and conditional variable can't be statically-initialized on Windows.
2145 static void init_threaded_search(void)
2147 init_recursive_mutex(&read_mutex
);
2148 pthread_mutex_init(&cache_mutex
, NULL
);
2149 pthread_mutex_init(&progress_mutex
, NULL
);
2150 pthread_cond_init(&progress_cond
, NULL
);
2151 old_try_to_free_routine
= set_try_to_free_routine(try_to_free_from_threads
);
2154 static void cleanup_threaded_search(void)
2156 set_try_to_free_routine(old_try_to_free_routine
);
2157 pthread_cond_destroy(&progress_cond
);
2158 pthread_mutex_destroy(&read_mutex
);
2159 pthread_mutex_destroy(&cache_mutex
);
2160 pthread_mutex_destroy(&progress_mutex
);
2163 static void *threaded_find_deltas(void *arg
)
2165 struct thread_params
*me
= arg
;
2167 while (me
->remaining
) {
2168 find_deltas(me
->list
, &me
->remaining
,
2169 me
->window
, me
->depth
, me
->processed
);
2173 pthread_cond_signal(&progress_cond
);
2177 * We must not set ->data_ready before we wait on the
2178 * condition because the main thread may have set it to 1
2179 * before we get here. In order to be sure that new
2180 * work is available if we see 1 in ->data_ready, it
2181 * was initialized to 0 before this thread was spawned
2182 * and we reset it to 0 right away.
2184 pthread_mutex_lock(&me
->mutex
);
2185 while (!me
->data_ready
)
2186 pthread_cond_wait(&me
->cond
, &me
->mutex
);
2188 pthread_mutex_unlock(&me
->mutex
);
2190 /* leave ->working 1 so that this doesn't get more work assigned */
2194 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
2195 int window
, int depth
, unsigned *processed
)
2197 struct thread_params
*p
;
2198 int i
, ret
, active_threads
= 0;
2200 init_threaded_search();
2202 if (delta_search_threads
<= 1) {
2203 find_deltas(list
, &list_size
, window
, depth
, processed
);
2204 cleanup_threaded_search();
2207 if (progress
> pack_to_stdout
)
2208 fprintf(stderr
, "Delta compression using up to %d threads.\n",
2209 delta_search_threads
);
2210 p
= xcalloc(delta_search_threads
, sizeof(*p
));
2212 /* Partition the work amongst work threads. */
2213 for (i
= 0; i
< delta_search_threads
; i
++) {
2214 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
2216 /* don't use too small segments or no deltas will be found */
2217 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
2220 p
[i
].window
= window
;
2222 p
[i
].processed
= processed
;
2224 p
[i
].data_ready
= 0;
2226 /* try to split chunks on "path" boundaries */
2227 while (sub_size
&& sub_size
< list_size
&&
2228 list
[sub_size
]->hash
&&
2229 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
2233 p
[i
].list_size
= sub_size
;
2234 p
[i
].remaining
= sub_size
;
2237 list_size
-= sub_size
;
2240 /* Start work threads. */
2241 for (i
= 0; i
< delta_search_threads
; i
++) {
2242 if (!p
[i
].list_size
)
2244 pthread_mutex_init(&p
[i
].mutex
, NULL
);
2245 pthread_cond_init(&p
[i
].cond
, NULL
);
2246 ret
= pthread_create(&p
[i
].thread
, NULL
,
2247 threaded_find_deltas
, &p
[i
]);
2249 die("unable to create thread: %s", strerror(ret
));
2254 * Now let's wait for work completion. Each time a thread is done
2255 * with its work, we steal half of the remaining work from the
2256 * thread with the largest number of unprocessed objects and give
2257 * it to that newly idle thread. This ensure good load balancing
2258 * until the remaining object list segments are simply too short
2259 * to be worth splitting anymore.
2261 while (active_threads
) {
2262 struct thread_params
*target
= NULL
;
2263 struct thread_params
*victim
= NULL
;
2264 unsigned sub_size
= 0;
2268 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
2273 pthread_cond_wait(&progress_cond
, &progress_mutex
);
2276 for (i
= 0; i
< delta_search_threads
; i
++)
2277 if (p
[i
].remaining
> 2*window
&&
2278 (!victim
|| victim
->remaining
< p
[i
].remaining
))
2281 sub_size
= victim
->remaining
/ 2;
2282 list
= victim
->list
+ victim
->list_size
- sub_size
;
2283 while (sub_size
&& list
[0]->hash
&&
2284 list
[0]->hash
== list
[-1]->hash
) {
2290 * It is possible for some "paths" to have
2291 * so many objects that no hash boundary
2292 * might be found. Let's just steal the
2293 * exact half in that case.
2295 sub_size
= victim
->remaining
/ 2;
2298 target
->list
= list
;
2299 victim
->list_size
-= sub_size
;
2300 victim
->remaining
-= sub_size
;
2302 target
->list_size
= sub_size
;
2303 target
->remaining
= sub_size
;
2304 target
->working
= 1;
2307 pthread_mutex_lock(&target
->mutex
);
2308 target
->data_ready
= 1;
2309 pthread_cond_signal(&target
->cond
);
2310 pthread_mutex_unlock(&target
->mutex
);
2313 pthread_join(target
->thread
, NULL
);
2314 pthread_cond_destroy(&target
->cond
);
2315 pthread_mutex_destroy(&target
->mutex
);
2319 cleanup_threaded_search();
2324 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2327 static void add_tag_chain(const struct object_id
*oid
)
2332 * We catch duplicates already in add_object_entry(), but we'd
2333 * prefer to do this extra check to avoid having to parse the
2334 * tag at all if we already know that it's being packed (e.g., if
2335 * it was included via bitmaps, we would not have parsed it
2338 if (packlist_find(&to_pack
, oid
->hash
, NULL
))
2341 tag
= lookup_tag(oid
->hash
);
2343 if (!tag
|| parse_tag(tag
) || !tag
->tagged
)
2344 die("unable to pack objects reachable from tag %s",
2347 add_object_entry(tag
->object
.oid
.hash
, OBJ_TAG
, NULL
, 0);
2349 if (tag
->tagged
->type
!= OBJ_TAG
)
2352 tag
= (struct tag
*)tag
->tagged
;
2356 static int add_ref_tag(const char *path
, const struct object_id
*oid
, int flag
, void *cb_data
)
2358 struct object_id peeled
;
2360 if (starts_with(path
, "refs/tags/") && /* is a tag? */
2361 !peel_ref(path
, peeled
.hash
) && /* peelable? */
2362 packlist_find(&to_pack
, peeled
.hash
, NULL
)) /* object packed? */
2367 static void prepare_pack(int window
, int depth
)
2369 struct object_entry
**delta_list
;
2370 uint32_t i
, nr_deltas
;
2373 get_object_details();
2376 * If we're locally repacking then we need to be doubly careful
2377 * from now on in order to make sure no stealth corruption gets
2378 * propagated to the new pack. Clients receiving streamed packs
2379 * should validate everything they get anyway so no need to incur
2380 * the additional cost here in that case.
2382 if (!pack_to_stdout
)
2383 do_check_packed_object_crc
= 1;
2385 if (!to_pack
.nr_objects
|| !window
|| !depth
)
2388 ALLOC_ARRAY(delta_list
, to_pack
.nr_objects
);
2391 for (i
= 0; i
< to_pack
.nr_objects
; i
++) {
2392 struct object_entry
*entry
= to_pack
.objects
+ i
;
2395 /* This happens if we decided to reuse existing
2396 * delta from a pack. "reuse_delta &&" is implied.
2400 if (entry
->size
< 50)
2403 if (entry
->no_try_delta
)
2406 if (!entry
->preferred_base
) {
2408 if (entry
->type
< 0)
2409 die("unable to get type of object %s",
2410 sha1_to_hex(entry
->idx
.sha1
));
2412 if (entry
->type
< 0) {
2414 * This object is not found, but we
2415 * don't have to include it anyway.
2421 delta_list
[n
++] = entry
;
2424 if (nr_deltas
&& n
> 1) {
2425 unsigned nr_done
= 0;
2427 progress_state
= start_progress(_("Compressing objects"),
2429 QSORT(delta_list
, n
, type_size_sort
);
2430 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
2431 stop_progress(&progress_state
);
2432 if (nr_done
!= nr_deltas
)
2433 die("inconsistency with delta count");
2438 static int git_pack_config(const char *k
, const char *v
, void *cb
)
2440 if (!strcmp(k
, "pack.window")) {
2441 window
= git_config_int(k
, v
);
2444 if (!strcmp(k
, "pack.windowmemory")) {
2445 window_memory_limit
= git_config_ulong(k
, v
);
2448 if (!strcmp(k
, "pack.depth")) {
2449 depth
= git_config_int(k
, v
);
2452 if (!strcmp(k
, "pack.deltacachesize")) {
2453 max_delta_cache_size
= git_config_int(k
, v
);
2456 if (!strcmp(k
, "pack.deltacachelimit")) {
2457 cache_max_small_delta_size
= git_config_int(k
, v
);
2460 if (!strcmp(k
, "pack.writebitmaphashcache")) {
2461 if (git_config_bool(k
, v
))
2462 write_bitmap_options
|= BITMAP_OPT_HASH_CACHE
;
2464 write_bitmap_options
&= ~BITMAP_OPT_HASH_CACHE
;
2466 if (!strcmp(k
, "pack.usebitmaps")) {
2467 use_bitmap_index_default
= git_config_bool(k
, v
);
2470 if (!strcmp(k
, "pack.threads")) {
2471 delta_search_threads
= git_config_int(k
, v
);
2472 if (delta_search_threads
< 0)
2473 die("invalid number of threads specified (%d)",
2474 delta_search_threads
);
2476 if (delta_search_threads
!= 1)
2477 warning("no threads support, ignoring %s", k
);
2481 if (!strcmp(k
, "pack.indexversion")) {
2482 pack_idx_opts
.version
= git_config_int(k
, v
);
2483 if (pack_idx_opts
.version
> 2)
2484 die("bad pack.indexversion=%"PRIu32
,
2485 pack_idx_opts
.version
);
2488 return git_default_config(k
, v
, cb
);
2491 static void read_object_list_from_stdin(void)
2493 char line
[40 + 1 + PATH_MAX
+ 2];
2494 unsigned char sha1
[20];
2497 if (!fgets(line
, sizeof(line
), stdin
)) {
2501 die("fgets returned NULL, not EOF, not error!");
2507 if (line
[0] == '-') {
2508 if (get_sha1_hex(line
+1, sha1
))
2509 die("expected edge sha1, got garbage:\n %s",
2511 add_preferred_base(sha1
);
2514 if (get_sha1_hex(line
, sha1
))
2515 die("expected sha1, got garbage:\n %s", line
);
2517 add_preferred_base_object(line
+41);
2518 add_object_entry(sha1
, 0, line
+41, 0);
2522 #define OBJECT_ADDED (1u<<20)
2524 static void show_commit(struct commit
*commit
, void *data
)
2526 add_object_entry(commit
->object
.oid
.hash
, OBJ_COMMIT
, NULL
, 0);
2527 commit
->object
.flags
|= OBJECT_ADDED
;
2529 if (write_bitmap_index
)
2530 index_commit_for_bitmap(commit
);
2533 static void show_object(struct object
*obj
, const char *name
, void *data
)
2535 add_preferred_base_object(name
);
2536 add_object_entry(obj
->oid
.hash
, obj
->type
, name
, 0);
2537 obj
->flags
|= OBJECT_ADDED
;
2540 static void show_edge(struct commit
*commit
)
2542 add_preferred_base(commit
->object
.oid
.hash
);
2545 struct in_pack_object
{
2547 struct object
*object
;
2553 struct in_pack_object
*array
;
2556 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
2558 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->oid
.hash
, p
);
2559 in_pack
->array
[in_pack
->nr
].object
= object
;
2564 * Compare the objects in the offset order, in order to emulate the
2565 * "git rev-list --objects" output that produced the pack originally.
2567 static int ofscmp(const void *a_
, const void *b_
)
2569 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
2570 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
2572 if (a
->offset
< b
->offset
)
2574 else if (a
->offset
> b
->offset
)
2577 return oidcmp(&a
->object
->oid
, &b
->object
->oid
);
2580 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
2582 struct packed_git
*p
;
2583 struct in_pack in_pack
;
2586 memset(&in_pack
, 0, sizeof(in_pack
));
2588 for (p
= packed_git
; p
; p
= p
->next
) {
2589 const unsigned char *sha1
;
2592 if (!p
->pack_local
|| p
->pack_keep
)
2594 if (open_pack_index(p
))
2595 die("cannot open pack index");
2597 ALLOC_GROW(in_pack
.array
,
2598 in_pack
.nr
+ p
->num_objects
,
2601 for (i
= 0; i
< p
->num_objects
; i
++) {
2602 sha1
= nth_packed_object_sha1(p
, i
);
2603 o
= lookup_unknown_object(sha1
);
2604 if (!(o
->flags
& OBJECT_ADDED
))
2605 mark_in_pack_object(o
, p
, &in_pack
);
2606 o
->flags
|= OBJECT_ADDED
;
2611 QSORT(in_pack
.array
, in_pack
.nr
, ofscmp
);
2612 for (i
= 0; i
< in_pack
.nr
; i
++) {
2613 struct object
*o
= in_pack
.array
[i
].object
;
2614 add_object_entry(o
->oid
.hash
, o
->type
, "", 0);
2617 free(in_pack
.array
);
2620 static int add_loose_object(const struct object_id
*oid
, const char *path
,
2623 enum object_type type
= sha1_object_info(oid
->hash
, NULL
);
2626 warning("loose object at %s could not be examined", path
);
2630 add_object_entry(oid
->hash
, type
, "", 0);
2635 * We actually don't even have to worry about reachability here.
2636 * add_object_entry will weed out duplicates, so we just add every
2637 * loose object we find.
2639 static void add_unreachable_loose_objects(void)
2641 for_each_loose_file_in_objdir(get_object_directory(),
2646 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1
)
2648 static struct packed_git
*last_found
= (void *)1;
2649 struct packed_git
*p
;
2651 p
= (last_found
!= (void *)1) ? last_found
: packed_git
;
2654 if ((!p
->pack_local
|| p
->pack_keep
) &&
2655 find_pack_entry_one(sha1
, p
)) {
2659 if (p
== last_found
)
2663 if (p
== last_found
)
2670 * Store a list of sha1s that are should not be discarded
2671 * because they are either written too recently, or are
2672 * reachable from another object that was.
2674 * This is filled by get_object_list.
2676 static struct oid_array recent_objects
;
2678 static int loosened_object_can_be_discarded(const struct object_id
*oid
,
2679 unsigned long mtime
)
2681 if (!unpack_unreachable_expiration
)
2683 if (mtime
> unpack_unreachable_expiration
)
2685 if (oid_array_lookup(&recent_objects
, oid
) >= 0)
2690 static void loosen_unused_packed_objects(struct rev_info
*revs
)
2692 struct packed_git
*p
;
2694 struct object_id oid
;
2696 for (p
= packed_git
; p
; p
= p
->next
) {
2697 if (!p
->pack_local
|| p
->pack_keep
)
2700 if (open_pack_index(p
))
2701 die("cannot open pack index");
2703 for (i
= 0; i
< p
->num_objects
; i
++) {
2704 nth_packed_object_oid(&oid
, p
, i
);
2705 if (!packlist_find(&to_pack
, oid
.hash
, NULL
) &&
2706 !has_sha1_pack_kept_or_nonlocal(oid
.hash
) &&
2707 !loosened_object_can_be_discarded(&oid
, p
->mtime
))
2708 if (force_object_loose(oid
.hash
, p
->mtime
))
2709 die("unable to force loose object");
2715 * This tracks any options which pack-reuse code expects to be on, or which a
2716 * reader of the pack might not understand, and which would therefore prevent
2717 * blind reuse of what we have on disk.
2719 static int pack_options_allow_reuse(void)
2721 return pack_to_stdout
&& allow_ofs_delta
;
2724 static int get_object_list_from_bitmap(struct rev_info
*revs
)
2726 if (prepare_bitmap_walk(revs
) < 0)
2729 if (pack_options_allow_reuse() &&
2730 !reuse_partial_packfile_from_bitmap(
2732 &reuse_packfile_objects
,
2733 &reuse_packfile_offset
)) {
2734 assert(reuse_packfile_objects
);
2735 nr_result
+= reuse_packfile_objects
;
2736 display_progress(progress_state
, nr_result
);
2739 traverse_bitmap_commit_list(&add_object_entry_from_bitmap
);
2743 static void record_recent_object(struct object
*obj
,
2747 oid_array_append(&recent_objects
, &obj
->oid
);
2750 static void record_recent_commit(struct commit
*commit
, void *data
)
2752 oid_array_append(&recent_objects
, &commit
->object
.oid
);
2755 static void get_object_list(int ac
, const char **av
)
2757 struct rev_info revs
;
2761 init_revisions(&revs
, NULL
);
2762 save_commit_buffer
= 0;
2763 setup_revisions(ac
, av
, &revs
, NULL
);
2765 /* make sure shallows are read */
2766 is_repository_shallow();
2768 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2769 int len
= strlen(line
);
2770 if (len
&& line
[len
- 1] == '\n')
2775 if (!strcmp(line
, "--not")) {
2776 flags
^= UNINTERESTING
;
2777 write_bitmap_index
= 0;
2780 if (starts_with(line
, "--shallow ")) {
2781 unsigned char sha1
[20];
2782 if (get_sha1_hex(line
+ 10, sha1
))
2783 die("not an SHA-1 '%s'", line
+ 10);
2784 register_shallow(sha1
);
2785 use_bitmap_index
= 0;
2788 die("not a rev '%s'", line
);
2790 if (handle_revision_arg(line
, &revs
, flags
, REVARG_CANNOT_BE_FILENAME
))
2791 die("bad revision '%s'", line
);
2794 if (use_bitmap_index
&& !get_object_list_from_bitmap(&revs
))
2797 if (prepare_revision_walk(&revs
))
2798 die("revision walk setup failed");
2799 mark_edges_uninteresting(&revs
, show_edge
);
2800 traverse_commit_list(&revs
, show_commit
, show_object
, NULL
);
2802 if (unpack_unreachable_expiration
) {
2803 revs
.ignore_missing_links
= 1;
2804 if (add_unseen_recent_objects_to_traversal(&revs
,
2805 unpack_unreachable_expiration
))
2806 die("unable to add recent objects");
2807 if (prepare_revision_walk(&revs
))
2808 die("revision walk setup failed");
2809 traverse_commit_list(&revs
, record_recent_commit
,
2810 record_recent_object
, NULL
);
2813 if (keep_unreachable
)
2814 add_objects_in_unpacked_packs(&revs
);
2815 if (pack_loose_unreachable
)
2816 add_unreachable_loose_objects();
2817 if (unpack_unreachable
)
2818 loosen_unused_packed_objects(&revs
);
2820 oid_array_clear(&recent_objects
);
2823 static int option_parse_index_version(const struct option
*opt
,
2824 const char *arg
, int unset
)
2827 const char *val
= arg
;
2828 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
2829 if (pack_idx_opts
.version
> 2)
2830 die(_("unsupported index version %s"), val
);
2831 if (*c
== ',' && c
[1])
2832 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
2833 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
2834 die(_("bad index version '%s'"), val
);
2838 static int option_parse_unpack_unreachable(const struct option
*opt
,
2839 const char *arg
, int unset
)
2842 unpack_unreachable
= 0;
2843 unpack_unreachable_expiration
= 0;
2846 unpack_unreachable
= 1;
2848 unpack_unreachable_expiration
= approxidate(arg
);
2853 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
2855 int use_internal_rev_list
= 0;
2858 int all_progress_implied
= 0;
2859 struct argv_array rp
= ARGV_ARRAY_INIT
;
2860 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
2861 int rev_list_index
= 0;
2862 struct option pack_objects_options
[] = {
2863 OPT_SET_INT('q', "quiet", &progress
,
2864 N_("do not show progress meter"), 0),
2865 OPT_SET_INT(0, "progress", &progress
,
2866 N_("show progress meter"), 1),
2867 OPT_SET_INT(0, "all-progress", &progress
,
2868 N_("show progress meter during object writing phase"), 2),
2869 OPT_BOOL(0, "all-progress-implied",
2870 &all_progress_implied
,
2871 N_("similar to --all-progress when progress meter is shown")),
2872 { OPTION_CALLBACK
, 0, "index-version", NULL
, N_("version[,offset]"),
2873 N_("write the pack index file in the specified idx format version"),
2874 0, option_parse_index_version
},
2875 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit
,
2876 N_("maximum size of each output pack file")),
2877 OPT_BOOL(0, "local", &local
,
2878 N_("ignore borrowed objects from alternate object store")),
2879 OPT_BOOL(0, "incremental", &incremental
,
2880 N_("ignore packed objects")),
2881 OPT_INTEGER(0, "window", &window
,
2882 N_("limit pack window by objects")),
2883 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit
,
2884 N_("limit pack window by memory in addition to object limit")),
2885 OPT_INTEGER(0, "depth", &depth
,
2886 N_("maximum length of delta chain allowed in the resulting pack")),
2887 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
2888 N_("reuse existing deltas")),
2889 OPT_BOOL(0, "reuse-object", &reuse_object
,
2890 N_("reuse existing objects")),
2891 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
2892 N_("use OFS_DELTA objects")),
2893 OPT_INTEGER(0, "threads", &delta_search_threads
,
2894 N_("use threads when searching for best delta matches")),
2895 OPT_BOOL(0, "non-empty", &non_empty
,
2896 N_("do not create an empty pack output")),
2897 OPT_BOOL(0, "revs", &use_internal_rev_list
,
2898 N_("read revision arguments from standard input")),
2899 { OPTION_SET_INT
, 0, "unpacked", &rev_list_unpacked
, NULL
,
2900 N_("limit the objects to those that are not yet packed"),
2901 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2902 { OPTION_SET_INT
, 0, "all", &rev_list_all
, NULL
,
2903 N_("include objects reachable from any reference"),
2904 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2905 { OPTION_SET_INT
, 0, "reflog", &rev_list_reflog
, NULL
,
2906 N_("include objects referred by reflog entries"),
2907 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2908 { OPTION_SET_INT
, 0, "indexed-objects", &rev_list_index
, NULL
,
2909 N_("include objects referred to by the index"),
2910 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2911 OPT_BOOL(0, "stdout", &pack_to_stdout
,
2912 N_("output pack to stdout")),
2913 OPT_BOOL(0, "include-tag", &include_tag
,
2914 N_("include tag objects that refer to objects to be packed")),
2915 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
2916 N_("keep unreachable objects")),
2917 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable
,
2918 N_("pack loose unreachable objects")),
2919 { OPTION_CALLBACK
, 0, "unpack-unreachable", NULL
, N_("time"),
2920 N_("unpack unreachable objects newer than <time>"),
2921 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
},
2922 OPT_BOOL(0, "thin", &thin
,
2923 N_("create thin packs")),
2924 OPT_BOOL(0, "shallow", &shallow
,
2925 N_("create packs suitable for shallow fetches")),
2926 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep
,
2927 N_("ignore packs that have companion .keep file")),
2928 OPT_INTEGER(0, "compression", &pack_compression_level
,
2929 N_("pack compression level")),
2930 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
2931 N_("do not hide commits by grafts"), 0),
2932 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index
,
2933 N_("use a bitmap index if available to speed up counting objects")),
2934 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index
,
2935 N_("write a bitmap index together with the pack index")),
2939 check_replace_refs
= 0;
2941 reset_pack_idx_option(&pack_idx_opts
);
2942 git_config(git_pack_config
, NULL
);
2944 progress
= isatty(2);
2945 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
2949 base_name
= argv
[0];
2952 if (pack_to_stdout
!= !base_name
|| argc
)
2953 usage_with_options(pack_usage
, pack_objects_options
);
2955 argv_array_push(&rp
, "pack-objects");
2957 use_internal_rev_list
= 1;
2958 argv_array_push(&rp
, shallow
2959 ? "--objects-edge-aggressive"
2960 : "--objects-edge");
2962 argv_array_push(&rp
, "--objects");
2965 use_internal_rev_list
= 1;
2966 argv_array_push(&rp
, "--all");
2968 if (rev_list_reflog
) {
2969 use_internal_rev_list
= 1;
2970 argv_array_push(&rp
, "--reflog");
2972 if (rev_list_index
) {
2973 use_internal_rev_list
= 1;
2974 argv_array_push(&rp
, "--indexed-objects");
2976 if (rev_list_unpacked
) {
2977 use_internal_rev_list
= 1;
2978 argv_array_push(&rp
, "--unpacked");
2983 if (pack_compression_level
== -1)
2984 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
2985 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
2986 die("bad pack compression level %d", pack_compression_level
);
2988 if (!delta_search_threads
) /* --threads=0 means autodetect */
2989 delta_search_threads
= online_cpus();
2992 if (delta_search_threads
!= 1)
2993 warning("no threads support, ignoring --threads");
2995 if (!pack_to_stdout
&& !pack_size_limit
)
2996 pack_size_limit
= pack_size_limit_cfg
;
2997 if (pack_to_stdout
&& pack_size_limit
)
2998 die("--max-pack-size cannot be used to build a pack for transfer.");
2999 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
3000 warning("minimum pack size limit is 1 MiB");
3001 pack_size_limit
= 1024*1024;
3004 if (!pack_to_stdout
&& thin
)
3005 die("--thin cannot be used to build an indexable pack.");
3007 if (keep_unreachable
&& unpack_unreachable
)
3008 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3009 if (!rev_list_all
|| !rev_list_reflog
|| !rev_list_index
)
3010 unpack_unreachable_expiration
= 0;
3013 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3015 * - to produce good pack (with bitmap index not-yet-packed objects are
3016 * packed in suboptimal order).
3018 * - to use more robust pack-generation codepath (avoiding possible
3019 * bugs in bitmap code and possible bitmap index corruption).
3021 if (!pack_to_stdout
)
3022 use_bitmap_index_default
= 0;
3024 if (use_bitmap_index
< 0)
3025 use_bitmap_index
= use_bitmap_index_default
;
3027 /* "hard" reasons not to use bitmaps; these just won't work at all */
3028 if (!use_internal_rev_list
|| (!pack_to_stdout
&& write_bitmap_index
) || is_repository_shallow())
3029 use_bitmap_index
= 0;
3031 if (pack_to_stdout
|| !rev_list_all
)
3032 write_bitmap_index
= 0;
3034 if (progress
&& all_progress_implied
)
3037 prepare_packed_git();
3038 if (ignore_packed_keep
) {
3039 struct packed_git
*p
;
3040 for (p
= packed_git
; p
; p
= p
->next
)
3041 if (p
->pack_local
&& p
->pack_keep
)
3043 if (!p
) /* no keep-able packs found */
3044 ignore_packed_keep
= 0;
3048 * unlike ignore_packed_keep above, we do not want to
3049 * unset "local" based on looking at packs, as it
3050 * also covers non-local objects
3052 struct packed_git
*p
;
3053 for (p
= packed_git
; p
; p
= p
->next
) {
3054 if (!p
->pack_local
) {
3055 have_non_local_packs
= 1;
3062 progress_state
= start_progress(_("Counting objects"), 0);
3063 if (!use_internal_rev_list
)
3064 read_object_list_from_stdin();
3066 get_object_list(rp
.argc
, rp
.argv
);
3067 argv_array_clear(&rp
);
3069 cleanup_preferred_base();
3070 if (include_tag
&& nr_result
)
3071 for_each_ref(add_ref_tag
, NULL
);
3072 stop_progress(&progress_state
);
3074 if (non_empty
&& !nr_result
)
3077 prepare_pack(window
, depth
);
3080 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
3081 " reused %"PRIu32
" (delta %"PRIu32
")\n",
3082 written
, written_delta
, reused
, reused_delta
);