11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
16 #include "list-objects.h"
19 #include "thread-utils.h"
21 static const char pack_usage
[] =
22 "git pack-objects [ -q | --progress | --all-progress ]\n"
23 " [--all-progress-implied]\n"
24 " [--max-pack-size=<n>] [--local] [--incremental]\n"
25 " [--window=<n>] [--window-memory=<n>] [--depth=<n>]\n"
26 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
27 " [--threads=<n>] [--non-empty] [--revs [--unpacked | --all]]\n"
28 " [--reflog] [--stdout | base-name] [--include-tag]\n"
29 " [--keep-unreachable | --unpack-unreachable]\n"
30 " [< ref-list | < object-list]";
33 struct pack_idx_entry idx
;
34 unsigned long size
; /* uncompressed size */
35 struct packed_git
*in_pack
; /* already in pack */
37 struct object_entry
*delta
; /* delta base object */
38 struct object_entry
*delta_child
; /* deltified objects who bases me */
39 struct object_entry
*delta_sibling
; /* other deltified objects who
40 * uses the same base as me
42 void *delta_data
; /* cached delta (uncompressed) */
43 unsigned long delta_size
; /* delta data size (uncompressed) */
44 unsigned long z_delta_size
; /* delta data size (compressed) */
45 unsigned int hash
; /* name hint hash */
46 enum object_type type
;
47 enum object_type in_pack_type
; /* could be delta */
48 unsigned char in_pack_header_size
;
49 unsigned char preferred_base
; /* we do not pack this, but is available
50 * to be used as the base object to delta
53 unsigned char no_try_delta
;
54 unsigned char tagged
; /* near the very tip of refs */
55 unsigned char filled
; /* assigned write-order */
59 * Objects we are going to pack are collected in objects array (dynamically
60 * expanded). nr_objects & nr_alloc controls this array. They are stored
61 * in the order we see -- typically rev-list --objects order that gives us
62 * nice "minimum seek" order.
64 static struct object_entry
*objects
;
65 static struct pack_idx_entry
**written_list
;
66 static uint32_t nr_objects
, nr_alloc
, nr_result
, nr_written
;
69 static int reuse_delta
= 1, reuse_object
= 1;
70 static int keep_unreachable
, unpack_unreachable
, include_tag
;
72 static int incremental
;
73 static int ignore_packed_keep
;
74 static int allow_ofs_delta
;
75 static struct pack_idx_option pack_idx_opts
;
76 static const char *base_name
;
77 static int progress
= 1;
78 static int window
= 10;
79 static unsigned long pack_size_limit
, pack_size_limit_cfg
;
80 static int depth
= 50;
81 static int delta_search_threads
;
82 static int pack_to_stdout
;
83 static int num_preferred_base
;
84 static struct progress
*progress_state
;
85 static int pack_compression_level
= Z_DEFAULT_COMPRESSION
;
86 static int pack_compression_seen
;
88 static unsigned long delta_cache_size
= 0;
89 static unsigned long max_delta_cache_size
= 256 * 1024 * 1024;
90 static unsigned long cache_max_small_delta_size
= 1000;
92 static unsigned long window_memory_limit
= 0;
95 * The object names in objects array are hashed with this hashtable,
96 * to help looking up the entry by object name.
97 * This hashtable is built after all the objects are seen.
99 static int *object_ix
;
100 static int object_ix_hashsz
;
101 static struct object_entry
*locate_object_entry(const unsigned char *sha1
);
106 static uint32_t written
, written_delta
;
107 static uint32_t reused
, reused_delta
;
110 static void *get_delta(struct object_entry
*entry
)
112 unsigned long size
, base_size
, delta_size
;
113 void *buf
, *base_buf
, *delta_buf
;
114 enum object_type type
;
116 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
118 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
119 base_buf
= read_sha1_file(entry
->delta
->idx
.sha1
, &type
, &base_size
);
121 die("unable to read %s", sha1_to_hex(entry
->delta
->idx
.sha1
));
122 delta_buf
= diff_delta(base_buf
, base_size
,
123 buf
, size
, &delta_size
, 0);
124 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
125 die("delta size changed");
131 static unsigned long do_compress(void **pptr
, unsigned long size
)
135 unsigned long maxsize
;
137 memset(&stream
, 0, sizeof(stream
));
138 git_deflate_init(&stream
, pack_compression_level
);
139 maxsize
= git_deflate_bound(&stream
, size
);
142 out
= xmalloc(maxsize
);
146 stream
.avail_in
= size
;
147 stream
.next_out
= out
;
148 stream
.avail_out
= maxsize
;
149 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
151 git_deflate_end(&stream
);
154 return stream
.total_out
;
158 * we are going to reuse the existing object data as is. make
159 * sure it is not corrupt.
161 static int check_pack_inflate(struct packed_git
*p
,
162 struct pack_window
**w_curs
,
165 unsigned long expect
)
168 unsigned char fakebuf
[4096], *in
;
171 memset(&stream
, 0, sizeof(stream
));
172 git_inflate_init(&stream
);
174 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
176 stream
.next_out
= fakebuf
;
177 stream
.avail_out
= sizeof(fakebuf
);
178 st
= git_inflate(&stream
, Z_FINISH
);
179 offset
+= stream
.next_in
- in
;
180 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
181 git_inflate_end(&stream
);
182 return (st
== Z_STREAM_END
&&
183 stream
.total_out
== expect
&&
184 stream
.total_in
== len
) ? 0 : -1;
187 static void copy_pack_data(struct sha1file
*f
,
188 struct packed_git
*p
,
189 struct pack_window
**w_curs
,
197 in
= use_pack(p
, w_curs
, offset
, &avail
);
199 avail
= (unsigned long)len
;
200 sha1write(f
, in
, avail
);
206 /* Return 0 if we will bust the pack-size limit */
207 static unsigned long write_object(struct sha1file
*f
,
208 struct object_entry
*entry
,
211 unsigned long size
, limit
, datalen
;
213 unsigned char header
[10], dheader
[10];
215 enum object_type type
;
216 int usable_delta
, to_reuse
;
223 /* apply size limit if limited packsize and not first object */
224 if (!pack_size_limit
|| !nr_written
)
226 else if (pack_size_limit
<= write_offset
)
228 * the earlier object did not fit the limit; avoid
229 * mistaking this with unlimited (i.e. limit = 0).
233 limit
= pack_size_limit
- write_offset
;
236 usable_delta
= 0; /* no delta */
237 else if (!pack_size_limit
)
238 usable_delta
= 1; /* unlimited packfile */
239 else if (entry
->delta
->idx
.offset
== (off_t
)-1)
240 usable_delta
= 0; /* base was written to another pack */
241 else if (entry
->delta
->idx
.offset
)
242 usable_delta
= 1; /* base already exists in this pack */
244 usable_delta
= 0; /* base could end up in another pack */
247 to_reuse
= 0; /* explicit */
248 else if (!entry
->in_pack
)
249 to_reuse
= 0; /* can't reuse what we don't have */
250 else if (type
== OBJ_REF_DELTA
|| type
== OBJ_OFS_DELTA
)
251 /* check_object() decided it for us ... */
252 to_reuse
= usable_delta
;
253 /* ... but pack split may override that */
254 else if (type
!= entry
->in_pack_type
)
255 to_reuse
= 0; /* pack has delta which is unusable */
256 else if (entry
->delta
)
257 to_reuse
= 0; /* we want to pack afresh */
259 to_reuse
= 1; /* we have it in-pack undeltified,
260 * and we do not need to deltify it.
266 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
268 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
270 * make sure no cached delta data remains from a
271 * previous attempt before a pack split occurred.
273 free(entry
->delta_data
);
274 entry
->delta_data
= NULL
;
275 entry
->z_delta_size
= 0;
276 } else if (entry
->delta_data
) {
277 size
= entry
->delta_size
;
278 buf
= entry
->delta_data
;
279 entry
->delta_data
= NULL
;
280 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
281 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
283 buf
= get_delta(entry
);
284 size
= entry
->delta_size
;
285 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
286 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
289 if (entry
->z_delta_size
)
290 datalen
= entry
->z_delta_size
;
292 datalen
= do_compress(&buf
, size
);
295 * The object header is a byte of 'type' followed by zero or
296 * more bytes of length.
298 hdrlen
= encode_in_pack_object_header(type
, size
, header
);
300 if (type
== OBJ_OFS_DELTA
) {
302 * Deltas with relative base contain an additional
303 * encoding of the relative offset for the delta
304 * base from this object's position in the pack.
306 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
307 unsigned pos
= sizeof(dheader
) - 1;
308 dheader
[pos
] = ofs
& 127;
310 dheader
[--pos
] = 128 | (--ofs
& 127);
311 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
315 sha1write(f
, header
, hdrlen
);
316 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
317 hdrlen
+= sizeof(dheader
) - pos
;
318 } else if (type
== OBJ_REF_DELTA
) {
320 * Deltas with a base reference contain
321 * an additional 20 bytes for the base sha1.
323 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
327 sha1write(f
, header
, hdrlen
);
328 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
331 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
335 sha1write(f
, header
, hdrlen
);
337 sha1write(f
, buf
, datalen
);
341 struct packed_git
*p
= entry
->in_pack
;
342 struct pack_window
*w_curs
= NULL
;
343 struct revindex_entry
*revidx
;
347 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
348 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
349 hdrlen
= encode_in_pack_object_header(type
, entry
->size
, header
);
351 offset
= entry
->in_pack_offset
;
352 revidx
= find_pack_revindex(p
, offset
);
353 datalen
= revidx
[1].offset
- offset
;
354 if (!pack_to_stdout
&& p
->index_version
> 1 &&
355 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
)) {
356 error("bad packed object CRC for %s", sha1_to_hex(entry
->idx
.sha1
));
361 offset
+= entry
->in_pack_header_size
;
362 datalen
-= entry
->in_pack_header_size
;
363 if (!pack_to_stdout
&& p
->index_version
== 1 &&
364 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry
->size
)) {
365 error("corrupt packed object for %s", sha1_to_hex(entry
->idx
.sha1
));
370 if (type
== OBJ_OFS_DELTA
) {
371 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
372 unsigned pos
= sizeof(dheader
) - 1;
373 dheader
[pos
] = ofs
& 127;
375 dheader
[--pos
] = 128 | (--ofs
& 127);
376 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
380 sha1write(f
, header
, hdrlen
);
381 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
382 hdrlen
+= sizeof(dheader
) - pos
;
384 } else if (type
== OBJ_REF_DELTA
) {
385 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
389 sha1write(f
, header
, hdrlen
);
390 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
394 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
398 sha1write(f
, header
, hdrlen
);
400 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
408 entry
->idx
.crc32
= crc32_end(f
);
409 return hdrlen
+ datalen
;
412 static int write_one(struct sha1file
*f
,
413 struct object_entry
*e
,
418 /* offset is non zero if object is written already. */
419 if (e
->idx
.offset
|| e
->preferred_base
)
422 /* if we are deltified, write out base object first. */
423 if (e
->delta
&& !write_one(f
, e
->delta
, offset
))
426 e
->idx
.offset
= *offset
;
427 size
= write_object(f
, e
, *offset
);
432 written_list
[nr_written
++] = &e
->idx
;
434 /* make sure off_t is sufficiently large not to wrap */
435 if (signed_add_overflows(*offset
, size
))
436 die("pack too large for current definition of off_t");
441 static int mark_tagged(const char *path
, const unsigned char *sha1
, int flag
,
444 unsigned char peeled
[20];
445 struct object_entry
*entry
= locate_object_entry(sha1
);
449 if (!peel_ref(path
, peeled
)) {
450 entry
= locate_object_entry(peeled
);
457 static void add_to_write_order(struct object_entry
**wo
,
459 struct object_entry
*e
)
467 static void add_descendants_to_write_order(struct object_entry
**wo
,
469 struct object_entry
*e
)
471 struct object_entry
*child
;
473 for (child
= e
->delta_child
; child
; child
= child
->delta_sibling
)
474 add_to_write_order(wo
, endp
, child
);
475 for (child
= e
->delta_child
; child
; child
= child
->delta_sibling
)
476 add_descendants_to_write_order(wo
, endp
, child
);
479 static void add_family_to_write_order(struct object_entry
**wo
,
481 struct object_entry
*e
)
483 struct object_entry
*root
;
485 for (root
= e
; root
->delta
; root
= root
->delta
)
487 add_to_write_order(wo
, endp
, root
);
488 add_descendants_to_write_order(wo
, endp
, root
);
491 static struct object_entry
**compute_write_order(void)
495 struct object_entry
**wo
= xmalloc(nr_objects
* sizeof(*wo
));
497 for (i
= 0; i
< nr_objects
; i
++) {
498 objects
[i
].tagged
= 0;
499 objects
[i
].filled
= 0;
500 objects
[i
].delta_child
= NULL
;
501 objects
[i
].delta_sibling
= NULL
;
505 * Fully connect delta_child/delta_sibling network.
506 * Make sure delta_sibling is sorted in the original
509 for (i
= nr_objects
- 1; 0 <= i
; i
--) {
510 struct object_entry
*e
= &objects
[i
];
513 /* Mark me as the first child */
514 e
->delta_sibling
= e
->delta
->delta_child
;
515 e
->delta
->delta_child
= e
;
519 * Mark objects that are at the tip of tags.
521 for_each_tag_ref(mark_tagged
, NULL
);
524 * Give the commits in the original recency order until
525 * we see a tagged tip.
527 for (i
= wo_end
= 0; i
< nr_objects
; i
++) {
528 if (objects
[i
].tagged
)
530 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
534 * Then fill all the tagged tips.
536 for (; i
< nr_objects
; i
++) {
537 if (objects
[i
].tagged
)
538 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
542 * And then all remaining commits and tags.
544 for (i
= 0; i
< nr_objects
; i
++) {
545 if (objects
[i
].type
!= OBJ_COMMIT
&&
546 objects
[i
].type
!= OBJ_TAG
)
548 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
552 * And then all the trees.
554 for (i
= 0; i
< nr_objects
; i
++) {
555 if (objects
[i
].type
!= OBJ_TREE
)
557 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
561 * Finally all the rest in really tight order
563 for (i
= 0; i
< nr_objects
; i
++)
564 add_family_to_write_order(wo
, &wo_end
, &objects
[i
]);
569 static void write_pack_file(void)
574 struct pack_header hdr
;
575 uint32_t nr_remaining
= nr_result
;
576 time_t last_mtime
= 0;
577 struct object_entry
**write_order
;
579 if (progress
> pack_to_stdout
)
580 progress_state
= start_progress("Writing objects", nr_result
);
581 written_list
= xmalloc(nr_objects
* sizeof(*written_list
));
582 write_order
= compute_write_order();
585 unsigned char sha1
[20];
586 char *pack_tmp_name
= NULL
;
588 if (pack_to_stdout
) {
589 f
= sha1fd_throughput(1, "<stdout>", progress_state
);
591 char tmpname
[PATH_MAX
];
593 fd
= odb_mkstemp(tmpname
, sizeof(tmpname
),
594 "pack/tmp_pack_XXXXXX");
595 pack_tmp_name
= xstrdup(tmpname
);
596 f
= sha1fd(fd
, pack_tmp_name
);
599 hdr
.hdr_signature
= htonl(PACK_SIGNATURE
);
600 hdr
.hdr_version
= htonl(PACK_VERSION
);
601 hdr
.hdr_entries
= htonl(nr_remaining
);
602 sha1write(f
, &hdr
, sizeof(hdr
));
603 offset
= sizeof(hdr
);
605 for (; i
< nr_objects
; i
++) {
606 struct object_entry
*e
= write_order
[i
];
607 if (!write_one(f
, e
, &offset
))
609 display_progress(progress_state
, written
);
613 * Did we write the wrong # entries in the header?
614 * If so, rewrite it like in fast-import
616 if (pack_to_stdout
) {
617 sha1close(f
, sha1
, CSUM_CLOSE
);
618 } else if (nr_written
== nr_remaining
) {
619 sha1close(f
, sha1
, CSUM_FSYNC
);
621 int fd
= sha1close(f
, sha1
, 0);
622 fixup_pack_header_footer(fd
, sha1
, pack_tmp_name
,
623 nr_written
, sha1
, offset
);
627 if (!pack_to_stdout
) {
629 const char *idx_tmp_name
;
630 char tmpname
[PATH_MAX
];
632 idx_tmp_name
= write_idx_file(NULL
, written_list
, nr_written
,
633 &pack_idx_opts
, sha1
);
635 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.pack",
636 base_name
, sha1_to_hex(sha1
));
637 free_pack_by_name(tmpname
);
638 if (adjust_shared_perm(pack_tmp_name
))
639 die_errno("unable to make temporary pack file readable");
640 if (rename(pack_tmp_name
, tmpname
))
641 die_errno("unable to rename temporary pack file");
644 * Packs are runtime accessed in their mtime
645 * order since newer packs are more likely to contain
646 * younger objects. So if we are creating multiple
647 * packs then we should modify the mtime of later ones
648 * to preserve this property.
650 if (stat(tmpname
, &st
) < 0) {
651 warning("failed to stat %s: %s",
652 tmpname
, strerror(errno
));
653 } else if (!last_mtime
) {
654 last_mtime
= st
.st_mtime
;
657 utb
.actime
= st
.st_atime
;
658 utb
.modtime
= --last_mtime
;
659 if (utime(tmpname
, &utb
) < 0)
660 warning("failed utime() on %s: %s",
661 tmpname
, strerror(errno
));
664 snprintf(tmpname
, sizeof(tmpname
), "%s-%s.idx",
665 base_name
, sha1_to_hex(sha1
));
666 if (adjust_shared_perm(idx_tmp_name
))
667 die_errno("unable to make temporary index file readable");
668 if (rename(idx_tmp_name
, tmpname
))
669 die_errno("unable to rename temporary index file");
671 free((void *) idx_tmp_name
);
673 puts(sha1_to_hex(sha1
));
676 /* mark written objects as written to previous pack */
677 for (j
= 0; j
< nr_written
; j
++) {
678 written_list
[j
]->offset
= (off_t
)-1;
680 nr_remaining
-= nr_written
;
681 } while (nr_remaining
&& i
< nr_objects
);
685 stop_progress(&progress_state
);
686 if (written
!= nr_result
)
687 die("wrote %"PRIu32
" objects while expecting %"PRIu32
,
691 static int locate_object_entry_hash(const unsigned char *sha1
)
695 memcpy(&ui
, sha1
, sizeof(unsigned int));
696 i
= ui
% object_ix_hashsz
;
697 while (0 < object_ix
[i
]) {
698 if (!hashcmp(sha1
, objects
[object_ix
[i
] - 1].idx
.sha1
))
700 if (++i
== object_ix_hashsz
)
706 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
710 if (!object_ix_hashsz
)
713 i
= locate_object_entry_hash(sha1
);
715 return &objects
[object_ix
[i
]-1];
719 static void rehash_objects(void)
722 struct object_entry
*oe
;
724 object_ix_hashsz
= nr_objects
* 3;
725 if (object_ix_hashsz
< 1024)
726 object_ix_hashsz
= 1024;
727 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
728 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
729 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
730 int ix
= locate_object_entry_hash(oe
->idx
.sha1
);
734 object_ix
[ix
] = i
+ 1;
738 static unsigned name_hash(const char *name
)
740 unsigned c
, hash
= 0;
746 * This effectively just creates a sortable number from the
747 * last sixteen non-whitespace characters. Last characters
748 * count "most", so things that end in ".c" sort together.
750 while ((c
= *name
++) != 0) {
753 hash
= (hash
>> 2) + (c
<< 24);
758 static void setup_delta_attr_check(struct git_attr_check
*check
)
760 static struct git_attr
*attr_delta
;
763 attr_delta
= git_attr("delta");
765 check
[0].attr
= attr_delta
;
768 static int no_try_delta(const char *path
)
770 struct git_attr_check check
[1];
772 setup_delta_attr_check(check
);
773 if (git_check_attr(path
, ARRAY_SIZE(check
), check
))
775 if (ATTR_FALSE(check
->value
))
780 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
781 const char *name
, int exclude
)
783 struct object_entry
*entry
;
784 struct packed_git
*p
, *found_pack
= NULL
;
785 off_t found_offset
= 0;
787 unsigned hash
= name_hash(name
);
789 ix
= nr_objects
? locate_object_entry_hash(sha1
) : -1;
792 entry
= objects
+ object_ix
[ix
] - 1;
793 if (!entry
->preferred_base
)
795 entry
->preferred_base
= 1;
800 if (!exclude
&& local
&& has_loose_object_nonlocal(sha1
))
803 for (p
= packed_git
; p
; p
= p
->next
) {
804 off_t offset
= find_pack_entry_one(sha1
, p
);
807 if (!is_pack_valid(p
)) {
808 warning("packfile %s cannot be accessed", p
->pack_name
);
811 found_offset
= offset
;
818 if (local
&& !p
->pack_local
)
820 if (ignore_packed_keep
&& p
->pack_local
&& p
->pack_keep
)
825 if (nr_objects
>= nr_alloc
) {
826 nr_alloc
= (nr_alloc
+ 1024) * 3 / 2;
827 objects
= xrealloc(objects
, nr_alloc
* sizeof(*entry
));
830 entry
= objects
+ nr_objects
++;
831 memset(entry
, 0, sizeof(*entry
));
832 hashcpy(entry
->idx
.sha1
, sha1
);
837 entry
->preferred_base
= 1;
841 entry
->in_pack
= found_pack
;
842 entry
->in_pack_offset
= found_offset
;
845 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
848 object_ix
[-1 - ix
] = nr_objects
;
850 display_progress(progress_state
, nr_objects
);
852 if (name
&& no_try_delta(name
))
853 entry
->no_try_delta
= 1;
858 struct pbase_tree_cache
{
859 unsigned char sha1
[20];
863 unsigned long tree_size
;
866 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
867 static int pbase_tree_cache_ix(const unsigned char *sha1
)
869 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
871 static int pbase_tree_cache_ix_incr(int ix
)
873 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
876 static struct pbase_tree
{
877 struct pbase_tree
*next
;
878 /* This is a phony "cache" entry; we are not
879 * going to evict it nor find it through _get()
880 * mechanism -- this is for the toplevel node that
881 * would almost always change with any commit.
883 struct pbase_tree_cache pcache
;
886 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
888 struct pbase_tree_cache
*ent
, *nent
;
891 enum object_type type
;
893 int my_ix
= pbase_tree_cache_ix(sha1
);
894 int available_ix
= -1;
896 /* pbase-tree-cache acts as a limited hashtable.
897 * your object will be found at your index or within a few
898 * slots after that slot if it is cached.
900 for (neigh
= 0; neigh
< 8; neigh
++) {
901 ent
= pbase_tree_cache
[my_ix
];
902 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
906 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
907 ((0 <= available_ix
) &&
908 (!ent
&& pbase_tree_cache
[available_ix
])))
909 available_ix
= my_ix
;
912 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
915 /* Did not find one. Either we got a bogus request or
916 * we need to read and perhaps cache.
918 data
= read_sha1_file(sha1
, &type
, &size
);
921 if (type
!= OBJ_TREE
) {
926 /* We need to either cache or return a throwaway copy */
928 if (available_ix
< 0)
931 ent
= pbase_tree_cache
[available_ix
];
932 my_ix
= available_ix
;
936 nent
= xmalloc(sizeof(*nent
));
937 nent
->temporary
= (available_ix
< 0);
940 /* evict and reuse */
941 free(ent
->tree_data
);
944 hashcpy(nent
->sha1
, sha1
);
945 nent
->tree_data
= data
;
946 nent
->tree_size
= size
;
948 if (!nent
->temporary
)
949 pbase_tree_cache
[my_ix
] = nent
;
953 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
955 if (!cache
->temporary
) {
959 free(cache
->tree_data
);
963 static int name_cmp_len(const char *name
)
966 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
971 static void add_pbase_object(struct tree_desc
*tree
,
974 const char *fullname
)
976 struct name_entry entry
;
979 while (tree_entry(tree
,&entry
)) {
980 if (S_ISGITLINK(entry
.mode
))
982 cmp
= tree_entry_len(entry
.path
, entry
.sha1
) != cmplen
? 1 :
983 memcmp(name
, entry
.path
, cmplen
);
988 if (name
[cmplen
] != '/') {
989 add_object_entry(entry
.sha1
,
990 object_type(entry
.mode
),
994 if (S_ISDIR(entry
.mode
)) {
995 struct tree_desc sub
;
996 struct pbase_tree_cache
*tree
;
997 const char *down
= name
+cmplen
+1;
998 int downlen
= name_cmp_len(down
);
1000 tree
= pbase_tree_get(entry
.sha1
);
1003 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1005 add_pbase_object(&sub
, down
, downlen
, fullname
);
1006 pbase_tree_put(tree
);
1011 static unsigned *done_pbase_paths
;
1012 static int done_pbase_paths_num
;
1013 static int done_pbase_paths_alloc
;
1014 static int done_pbase_path_pos(unsigned hash
)
1017 int hi
= done_pbase_paths_num
;
1019 int mi
= (hi
+ lo
) / 2;
1020 if (done_pbase_paths
[mi
] == hash
)
1022 if (done_pbase_paths
[mi
] < hash
)
1030 static int check_pbase_path(unsigned hash
)
1032 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
1036 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
1037 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
1038 done_pbase_paths
= xrealloc(done_pbase_paths
,
1039 done_pbase_paths_alloc
*
1042 done_pbase_paths_num
++;
1043 if (pos
< done_pbase_paths_num
)
1044 memmove(done_pbase_paths
+ pos
+ 1,
1045 done_pbase_paths
+ pos
,
1046 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
1047 done_pbase_paths
[pos
] = hash
;
1051 static void add_preferred_base_object(const char *name
)
1053 struct pbase_tree
*it
;
1055 unsigned hash
= name_hash(name
);
1057 if (!num_preferred_base
|| check_pbase_path(hash
))
1060 cmplen
= name_cmp_len(name
);
1061 for (it
= pbase_tree
; it
; it
= it
->next
) {
1063 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
1066 struct tree_desc tree
;
1067 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1068 add_pbase_object(&tree
, name
, cmplen
, name
);
1073 static void add_preferred_base(unsigned char *sha1
)
1075 struct pbase_tree
*it
;
1078 unsigned char tree_sha1
[20];
1080 if (window
<= num_preferred_base
++)
1083 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
1087 for (it
= pbase_tree
; it
; it
= it
->next
) {
1088 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
1094 it
= xcalloc(1, sizeof(*it
));
1095 it
->next
= pbase_tree
;
1098 hashcpy(it
->pcache
.sha1
, tree_sha1
);
1099 it
->pcache
.tree_data
= data
;
1100 it
->pcache
.tree_size
= size
;
1103 static void cleanup_preferred_base(void)
1105 struct pbase_tree
*it
;
1111 struct pbase_tree
*this = it
;
1113 free(this->pcache
.tree_data
);
1117 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1118 if (!pbase_tree_cache
[i
])
1120 free(pbase_tree_cache
[i
]->tree_data
);
1121 free(pbase_tree_cache
[i
]);
1122 pbase_tree_cache
[i
] = NULL
;
1125 free(done_pbase_paths
);
1126 done_pbase_paths
= NULL
;
1127 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1130 static void check_object(struct object_entry
*entry
)
1132 if (entry
->in_pack
) {
1133 struct packed_git
*p
= entry
->in_pack
;
1134 struct pack_window
*w_curs
= NULL
;
1135 const unsigned char *base_ref
= NULL
;
1136 struct object_entry
*base_entry
;
1137 unsigned long used
, used_0
;
1138 unsigned long avail
;
1140 unsigned char *buf
, c
;
1142 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1145 * We want in_pack_type even if we do not reuse delta
1146 * since non-delta representations could still be reused.
1148 used
= unpack_object_header_buffer(buf
, avail
,
1149 &entry
->in_pack_type
,
1155 * Determine if this is a delta and if so whether we can
1156 * reuse it or not. Otherwise let's find out as cheaply as
1157 * possible what the actual type and size for this object is.
1159 switch (entry
->in_pack_type
) {
1161 /* Not a delta hence we've already got all we need. */
1162 entry
->type
= entry
->in_pack_type
;
1163 entry
->in_pack_header_size
= used
;
1164 if (entry
->type
< OBJ_COMMIT
|| entry
->type
> OBJ_BLOB
)
1166 unuse_pack(&w_curs
);
1169 if (reuse_delta
&& !entry
->preferred_base
)
1170 base_ref
= use_pack(p
, &w_curs
,
1171 entry
->in_pack_offset
+ used
, NULL
);
1172 entry
->in_pack_header_size
= used
+ 20;
1175 buf
= use_pack(p
, &w_curs
,
1176 entry
->in_pack_offset
+ used
, NULL
);
1182 if (!ofs
|| MSB(ofs
, 7)) {
1183 error("delta base offset overflow in pack for %s",
1184 sha1_to_hex(entry
->idx
.sha1
));
1188 ofs
= (ofs
<< 7) + (c
& 127);
1190 ofs
= entry
->in_pack_offset
- ofs
;
1191 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1192 error("delta base offset out of bound for %s",
1193 sha1_to_hex(entry
->idx
.sha1
));
1196 if (reuse_delta
&& !entry
->preferred_base
) {
1197 struct revindex_entry
*revidx
;
1198 revidx
= find_pack_revindex(p
, ofs
);
1201 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1203 entry
->in_pack_header_size
= used
+ used_0
;
1207 if (base_ref
&& (base_entry
= locate_object_entry(base_ref
))) {
1209 * If base_ref was set above that means we wish to
1210 * reuse delta data, and we even found that base
1211 * in the list of objects we want to pack. Goodie!
1213 * Depth value does not matter - find_deltas() will
1214 * never consider reused delta as the base object to
1215 * deltify other objects against, in order to avoid
1218 entry
->type
= entry
->in_pack_type
;
1219 entry
->delta
= base_entry
;
1220 entry
->delta_size
= entry
->size
;
1221 entry
->delta_sibling
= base_entry
->delta_child
;
1222 base_entry
->delta_child
= entry
;
1223 unuse_pack(&w_curs
);
1229 * This must be a delta and we already know what the
1230 * final object type is. Let's extract the actual
1231 * object size from the delta header.
1233 entry
->size
= get_size_from_delta(p
, &w_curs
,
1234 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1235 if (entry
->size
== 0)
1237 unuse_pack(&w_curs
);
1242 * No choice but to fall back to the recursive delta walk
1243 * with sha1_object_info() to find about the object type
1247 unuse_pack(&w_curs
);
1250 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1252 * The error condition is checked in prepare_pack(). This is
1253 * to permit a missing preferred base object to be ignored
1254 * as a preferred base. Doing so can result in a larger
1255 * pack file, but the transfer will still take place.
1259 static int pack_offset_sort(const void *_a
, const void *_b
)
1261 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1262 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1264 /* avoid filesystem trashing with loose objects */
1265 if (!a
->in_pack
&& !b
->in_pack
)
1266 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1268 if (a
->in_pack
< b
->in_pack
)
1270 if (a
->in_pack
> b
->in_pack
)
1272 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1273 (a
->in_pack_offset
> b
->in_pack_offset
);
1276 static void get_object_details(void)
1279 struct object_entry
**sorted_by_offset
;
1281 sorted_by_offset
= xcalloc(nr_objects
, sizeof(struct object_entry
*));
1282 for (i
= 0; i
< nr_objects
; i
++)
1283 sorted_by_offset
[i
] = objects
+ i
;
1284 qsort(sorted_by_offset
, nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1286 for (i
= 0; i
< nr_objects
; i
++) {
1287 struct object_entry
*entry
= sorted_by_offset
[i
];
1288 check_object(entry
);
1289 if (big_file_threshold
<= entry
->size
)
1290 entry
->no_try_delta
= 1;
1293 free(sorted_by_offset
);
1297 * We search for deltas in a list sorted by type, by filename hash, and then
1298 * by size, so that we see progressively smaller and smaller files.
1299 * That's because we prefer deltas to be from the bigger file
1300 * to the smaller -- deletes are potentially cheaper, but perhaps
1301 * more importantly, the bigger file is likely the more recent
1302 * one. The deepest deltas are therefore the oldest objects which are
1303 * less susceptible to be accessed often.
1305 static int type_size_sort(const void *_a
, const void *_b
)
1307 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1308 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1310 if (a
->type
> b
->type
)
1312 if (a
->type
< b
->type
)
1314 if (a
->hash
> b
->hash
)
1316 if (a
->hash
< b
->hash
)
1318 if (a
->preferred_base
> b
->preferred_base
)
1320 if (a
->preferred_base
< b
->preferred_base
)
1322 if (a
->size
> b
->size
)
1324 if (a
->size
< b
->size
)
1326 return a
< b
? -1 : (a
> b
); /* newest first */
1330 struct object_entry
*entry
;
1332 struct delta_index
*index
;
1336 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1337 unsigned long delta_size
)
1339 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1342 if (delta_size
< cache_max_small_delta_size
)
1345 /* cache delta, if objects are large enough compared to delta size */
1346 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1354 static pthread_mutex_t read_mutex
;
1355 #define read_lock() pthread_mutex_lock(&read_mutex)
1356 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1358 static pthread_mutex_t cache_mutex
;
1359 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1360 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1362 static pthread_mutex_t progress_mutex
;
1363 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1364 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1368 #define read_lock() (void)0
1369 #define read_unlock() (void)0
1370 #define cache_lock() (void)0
1371 #define cache_unlock() (void)0
1372 #define progress_lock() (void)0
1373 #define progress_unlock() (void)0
1377 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1378 unsigned max_depth
, unsigned long *mem_usage
)
1380 struct object_entry
*trg_entry
= trg
->entry
;
1381 struct object_entry
*src_entry
= src
->entry
;
1382 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1384 enum object_type type
;
1387 /* Don't bother doing diffs between different types */
1388 if (trg_entry
->type
!= src_entry
->type
)
1392 * We do not bother to try a delta that we discarded
1393 * on an earlier try, but only when reusing delta data.
1395 if (reuse_delta
&& trg_entry
->in_pack
&&
1396 trg_entry
->in_pack
== src_entry
->in_pack
&&
1397 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1398 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1401 /* Let's not bust the allowed depth. */
1402 if (src
->depth
>= max_depth
)
1405 /* Now some size filtering heuristics. */
1406 trg_size
= trg_entry
->size
;
1407 if (!trg_entry
->delta
) {
1408 max_size
= trg_size
/2 - 20;
1411 max_size
= trg_entry
->delta_size
;
1412 ref_depth
= trg
->depth
;
1414 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
1415 (max_depth
- ref_depth
+ 1);
1418 src_size
= src_entry
->size
;
1419 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1420 if (sizediff
>= max_size
)
1422 if (trg_size
< src_size
/ 32)
1425 /* Load data if not already done */
1428 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1431 die("object %s cannot be read",
1432 sha1_to_hex(trg_entry
->idx
.sha1
));
1434 die("object %s inconsistent object length (%lu vs %lu)",
1435 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1440 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1443 if (src_entry
->preferred_base
) {
1444 static int warned
= 0;
1446 warning("object %s cannot be read",
1447 sha1_to_hex(src_entry
->idx
.sha1
));
1449 * Those objects are not included in the
1450 * resulting pack. Be resilient and ignore
1451 * them if they can't be read, in case the
1452 * pack could be created nevertheless.
1456 die("object %s cannot be read",
1457 sha1_to_hex(src_entry
->idx
.sha1
));
1460 die("object %s inconsistent object length (%lu vs %lu)",
1461 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1465 src
->index
= create_delta_index(src
->data
, src_size
);
1467 static int warned
= 0;
1469 warning("suboptimal pack - out of memory");
1472 *mem_usage
+= sizeof_delta_index(src
->index
);
1475 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1479 if (trg_entry
->delta
) {
1480 /* Prefer only shallower same-sized deltas. */
1481 if (delta_size
== trg_entry
->delta_size
&&
1482 src
->depth
+ 1 >= trg
->depth
) {
1489 * Handle memory allocation outside of the cache
1490 * accounting lock. Compiler will optimize the strangeness
1491 * away when NO_PTHREADS is defined.
1493 free(trg_entry
->delta_data
);
1495 if (trg_entry
->delta_data
) {
1496 delta_cache_size
-= trg_entry
->delta_size
;
1497 trg_entry
->delta_data
= NULL
;
1499 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1500 delta_cache_size
+= delta_size
;
1502 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1508 trg_entry
->delta
= src_entry
;
1509 trg_entry
->delta_size
= delta_size
;
1510 trg
->depth
= src
->depth
+ 1;
1515 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1517 struct object_entry
*child
= me
->delta_child
;
1520 unsigned int c
= check_delta_limit(child
, n
+ 1);
1523 child
= child
->delta_sibling
;
1528 static unsigned long free_unpacked(struct unpacked
*n
)
1530 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1531 free_delta_index(n
->index
);
1534 freed_mem
+= n
->entry
->size
;
1543 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1544 int window
, int depth
, unsigned *processed
)
1546 uint32_t i
, idx
= 0, count
= 0;
1547 struct unpacked
*array
;
1548 unsigned long mem_usage
= 0;
1550 array
= xcalloc(window
, sizeof(struct unpacked
));
1553 struct object_entry
*entry
;
1554 struct unpacked
*n
= array
+ idx
;
1555 int j
, max_depth
, best_base
= -1;
1564 if (!entry
->preferred_base
) {
1566 display_progress(progress_state
, *processed
);
1570 mem_usage
-= free_unpacked(n
);
1573 while (window_memory_limit
&&
1574 mem_usage
> window_memory_limit
&&
1576 uint32_t tail
= (idx
+ window
- count
) % window
;
1577 mem_usage
-= free_unpacked(array
+ tail
);
1581 /* We do not compute delta to *create* objects we are not
1584 if (entry
->preferred_base
)
1588 * If the current object is at pack edge, take the depth the
1589 * objects that depend on the current object into account
1590 * otherwise they would become too deep.
1593 if (entry
->delta_child
) {
1594 max_depth
-= check_delta_limit(entry
, 0);
1602 uint32_t other_idx
= idx
+ j
;
1604 if (other_idx
>= window
)
1605 other_idx
-= window
;
1606 m
= array
+ other_idx
;
1609 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1613 best_base
= other_idx
;
1617 * If we decided to cache the delta data, then it is best
1618 * to compress it right away. First because we have to do
1619 * it anyway, and doing it here while we're threaded will
1620 * save a lot of time in the non threaded write phase,
1621 * as well as allow for caching more deltas within
1622 * the same cache size limit.
1624 * But only if not writing to stdout, since in that case
1625 * the network is most likely throttling writes anyway,
1626 * and therefore it is best to go to the write phase ASAP
1627 * instead, as we can afford spending more time compressing
1628 * between writes at that moment.
1630 if (entry
->delta_data
&& !pack_to_stdout
) {
1631 entry
->z_delta_size
= do_compress(&entry
->delta_data
,
1634 delta_cache_size
-= entry
->delta_size
;
1635 delta_cache_size
+= entry
->z_delta_size
;
1639 /* if we made n a delta, and if n is already at max
1640 * depth, leaving it in the window is pointless. we
1641 * should evict it first.
1643 if (entry
->delta
&& max_depth
<= n
->depth
)
1647 * Move the best delta base up in the window, after the
1648 * currently deltified object, to keep it longer. It will
1649 * be the first base object to be attempted next.
1652 struct unpacked swap
= array
[best_base
];
1653 int dist
= (window
+ idx
- best_base
) % window
;
1654 int dst
= best_base
;
1656 int src
= (dst
+ 1) % window
;
1657 array
[dst
] = array
[src
];
1665 if (count
+ 1 < window
)
1671 for (i
= 0; i
< window
; ++i
) {
1672 free_delta_index(array
[i
].index
);
1673 free(array
[i
].data
);
1680 static void try_to_free_from_threads(size_t size
)
1683 release_pack_memory(size
, -1);
1687 static try_to_free_t old_try_to_free_routine
;
1690 * The main thread waits on the condition that (at least) one of the workers
1691 * has stopped working (which is indicated in the .working member of
1692 * struct thread_params).
1693 * When a work thread has completed its work, it sets .working to 0 and
1694 * signals the main thread and waits on the condition that .data_ready
1698 struct thread_params
{
1700 struct object_entry
**list
;
1707 pthread_mutex_t mutex
;
1708 pthread_cond_t cond
;
1709 unsigned *processed
;
1712 static pthread_cond_t progress_cond
;
1715 * Mutex and conditional variable can't be statically-initialized on Windows.
1717 static void init_threaded_search(void)
1719 init_recursive_mutex(&read_mutex
);
1720 pthread_mutex_init(&cache_mutex
, NULL
);
1721 pthread_mutex_init(&progress_mutex
, NULL
);
1722 pthread_cond_init(&progress_cond
, NULL
);
1723 old_try_to_free_routine
= set_try_to_free_routine(try_to_free_from_threads
);
1726 static void cleanup_threaded_search(void)
1728 set_try_to_free_routine(old_try_to_free_routine
);
1729 pthread_cond_destroy(&progress_cond
);
1730 pthread_mutex_destroy(&read_mutex
);
1731 pthread_mutex_destroy(&cache_mutex
);
1732 pthread_mutex_destroy(&progress_mutex
);
1735 static void *threaded_find_deltas(void *arg
)
1737 struct thread_params
*me
= arg
;
1739 while (me
->remaining
) {
1740 find_deltas(me
->list
, &me
->remaining
,
1741 me
->window
, me
->depth
, me
->processed
);
1745 pthread_cond_signal(&progress_cond
);
1749 * We must not set ->data_ready before we wait on the
1750 * condition because the main thread may have set it to 1
1751 * before we get here. In order to be sure that new
1752 * work is available if we see 1 in ->data_ready, it
1753 * was initialized to 0 before this thread was spawned
1754 * and we reset it to 0 right away.
1756 pthread_mutex_lock(&me
->mutex
);
1757 while (!me
->data_ready
)
1758 pthread_cond_wait(&me
->cond
, &me
->mutex
);
1760 pthread_mutex_unlock(&me
->mutex
);
1762 /* leave ->working 1 so that this doesn't get more work assigned */
1766 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
1767 int window
, int depth
, unsigned *processed
)
1769 struct thread_params
*p
;
1770 int i
, ret
, active_threads
= 0;
1772 init_threaded_search();
1774 if (!delta_search_threads
) /* --threads=0 means autodetect */
1775 delta_search_threads
= online_cpus();
1776 if (delta_search_threads
<= 1) {
1777 find_deltas(list
, &list_size
, window
, depth
, processed
);
1778 cleanup_threaded_search();
1781 if (progress
> pack_to_stdout
)
1782 fprintf(stderr
, "Delta compression using up to %d threads.\n",
1783 delta_search_threads
);
1784 p
= xcalloc(delta_search_threads
, sizeof(*p
));
1786 /* Partition the work amongst work threads. */
1787 for (i
= 0; i
< delta_search_threads
; i
++) {
1788 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
1790 /* don't use too small segments or no deltas will be found */
1791 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
1794 p
[i
].window
= window
;
1796 p
[i
].processed
= processed
;
1798 p
[i
].data_ready
= 0;
1800 /* try to split chunks on "path" boundaries */
1801 while (sub_size
&& sub_size
< list_size
&&
1802 list
[sub_size
]->hash
&&
1803 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
1807 p
[i
].list_size
= sub_size
;
1808 p
[i
].remaining
= sub_size
;
1811 list_size
-= sub_size
;
1814 /* Start work threads. */
1815 for (i
= 0; i
< delta_search_threads
; i
++) {
1816 if (!p
[i
].list_size
)
1818 pthread_mutex_init(&p
[i
].mutex
, NULL
);
1819 pthread_cond_init(&p
[i
].cond
, NULL
);
1820 ret
= pthread_create(&p
[i
].thread
, NULL
,
1821 threaded_find_deltas
, &p
[i
]);
1823 die("unable to create thread: %s", strerror(ret
));
1828 * Now let's wait for work completion. Each time a thread is done
1829 * with its work, we steal half of the remaining work from the
1830 * thread with the largest number of unprocessed objects and give
1831 * it to that newly idle thread. This ensure good load balancing
1832 * until the remaining object list segments are simply too short
1833 * to be worth splitting anymore.
1835 while (active_threads
) {
1836 struct thread_params
*target
= NULL
;
1837 struct thread_params
*victim
= NULL
;
1838 unsigned sub_size
= 0;
1842 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
1847 pthread_cond_wait(&progress_cond
, &progress_mutex
);
1850 for (i
= 0; i
< delta_search_threads
; i
++)
1851 if (p
[i
].remaining
> 2*window
&&
1852 (!victim
|| victim
->remaining
< p
[i
].remaining
))
1855 sub_size
= victim
->remaining
/ 2;
1856 list
= victim
->list
+ victim
->list_size
- sub_size
;
1857 while (sub_size
&& list
[0]->hash
&&
1858 list
[0]->hash
== list
[-1]->hash
) {
1864 * It is possible for some "paths" to have
1865 * so many objects that no hash boundary
1866 * might be found. Let's just steal the
1867 * exact half in that case.
1869 sub_size
= victim
->remaining
/ 2;
1872 target
->list
= list
;
1873 victim
->list_size
-= sub_size
;
1874 victim
->remaining
-= sub_size
;
1876 target
->list_size
= sub_size
;
1877 target
->remaining
= sub_size
;
1878 target
->working
= 1;
1881 pthread_mutex_lock(&target
->mutex
);
1882 target
->data_ready
= 1;
1883 pthread_cond_signal(&target
->cond
);
1884 pthread_mutex_unlock(&target
->mutex
);
1887 pthread_join(target
->thread
, NULL
);
1888 pthread_cond_destroy(&target
->cond
);
1889 pthread_mutex_destroy(&target
->mutex
);
1893 cleanup_threaded_search();
1898 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1901 static int add_ref_tag(const char *path
, const unsigned char *sha1
, int flag
, void *cb_data
)
1903 unsigned char peeled
[20];
1905 if (!prefixcmp(path
, "refs/tags/") && /* is a tag? */
1906 !peel_ref(path
, peeled
) && /* peelable? */
1907 !is_null_sha1(peeled
) && /* annotated tag? */
1908 locate_object_entry(peeled
)) /* object packed? */
1909 add_object_entry(sha1
, OBJ_TAG
, NULL
, 0);
1913 static void prepare_pack(int window
, int depth
)
1915 struct object_entry
**delta_list
;
1916 uint32_t i
, nr_deltas
;
1919 get_object_details();
1922 * If we're locally repacking then we need to be doubly careful
1923 * from now on in order to make sure no stealth corruption gets
1924 * propagated to the new pack. Clients receiving streamed packs
1925 * should validate everything they get anyway so no need to incur
1926 * the additional cost here in that case.
1928 if (!pack_to_stdout
)
1929 do_check_packed_object_crc
= 1;
1931 if (!nr_objects
|| !window
|| !depth
)
1934 delta_list
= xmalloc(nr_objects
* sizeof(*delta_list
));
1937 for (i
= 0; i
< nr_objects
; i
++) {
1938 struct object_entry
*entry
= objects
+ i
;
1941 /* This happens if we decided to reuse existing
1942 * delta from a pack. "reuse_delta &&" is implied.
1946 if (entry
->size
< 50)
1949 if (entry
->no_try_delta
)
1952 if (!entry
->preferred_base
) {
1954 if (entry
->type
< 0)
1955 die("unable to get type of object %s",
1956 sha1_to_hex(entry
->idx
.sha1
));
1958 if (entry
->type
< 0) {
1960 * This object is not found, but we
1961 * don't have to include it anyway.
1967 delta_list
[n
++] = entry
;
1970 if (nr_deltas
&& n
> 1) {
1971 unsigned nr_done
= 0;
1973 progress_state
= start_progress("Compressing objects",
1975 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
1976 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
1977 stop_progress(&progress_state
);
1978 if (nr_done
!= nr_deltas
)
1979 die("inconsistency with delta count");
1984 static int git_pack_config(const char *k
, const char *v
, void *cb
)
1986 if (!strcmp(k
, "pack.window")) {
1987 window
= git_config_int(k
, v
);
1990 if (!strcmp(k
, "pack.windowmemory")) {
1991 window_memory_limit
= git_config_ulong(k
, v
);
1994 if (!strcmp(k
, "pack.depth")) {
1995 depth
= git_config_int(k
, v
);
1998 if (!strcmp(k
, "pack.compression")) {
1999 int level
= git_config_int(k
, v
);
2001 level
= Z_DEFAULT_COMPRESSION
;
2002 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
2003 die("bad pack compression level %d", level
);
2004 pack_compression_level
= level
;
2005 pack_compression_seen
= 1;
2008 if (!strcmp(k
, "pack.deltacachesize")) {
2009 max_delta_cache_size
= git_config_int(k
, v
);
2012 if (!strcmp(k
, "pack.deltacachelimit")) {
2013 cache_max_small_delta_size
= git_config_int(k
, v
);
2016 if (!strcmp(k
, "pack.threads")) {
2017 delta_search_threads
= git_config_int(k
, v
);
2018 if (delta_search_threads
< 0)
2019 die("invalid number of threads specified (%d)",
2020 delta_search_threads
);
2022 if (delta_search_threads
!= 1)
2023 warning("no threads support, ignoring %s", k
);
2027 if (!strcmp(k
, "pack.indexversion")) {
2028 pack_idx_opts
.version
= git_config_int(k
, v
);
2029 if (pack_idx_opts
.version
> 2)
2030 die("bad pack.indexversion=%"PRIu32
,
2031 pack_idx_opts
.version
);
2034 if (!strcmp(k
, "pack.packsizelimit")) {
2035 pack_size_limit_cfg
= git_config_ulong(k
, v
);
2038 return git_default_config(k
, v
, cb
);
2041 static void read_object_list_from_stdin(void)
2043 char line
[40 + 1 + PATH_MAX
+ 2];
2044 unsigned char sha1
[20];
2047 if (!fgets(line
, sizeof(line
), stdin
)) {
2051 die("fgets returned NULL, not EOF, not error!");
2057 if (line
[0] == '-') {
2058 if (get_sha1_hex(line
+1, sha1
))
2059 die("expected edge sha1, got garbage:\n %s",
2061 add_preferred_base(sha1
);
2064 if (get_sha1_hex(line
, sha1
))
2065 die("expected sha1, got garbage:\n %s", line
);
2067 add_preferred_base_object(line
+41);
2068 add_object_entry(sha1
, 0, line
+41, 0);
2072 #define OBJECT_ADDED (1u<<20)
2074 static void show_commit(struct commit
*commit
, void *data
)
2076 add_object_entry(commit
->object
.sha1
, OBJ_COMMIT
, NULL
, 0);
2077 commit
->object
.flags
|= OBJECT_ADDED
;
2080 static void show_object(struct object
*obj
,
2081 const struct name_path
*path
, const char *last
,
2084 char *name
= path_name(path
, last
);
2086 add_preferred_base_object(name
);
2087 add_object_entry(obj
->sha1
, obj
->type
, name
, 0);
2088 obj
->flags
|= OBJECT_ADDED
;
2091 * We will have generated the hash from the name,
2092 * but not saved a pointer to it - we can free it
2097 static void show_edge(struct commit
*commit
)
2099 add_preferred_base(commit
->object
.sha1
);
2102 struct in_pack_object
{
2104 struct object
*object
;
2110 struct in_pack_object
*array
;
2113 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
2115 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->sha1
, p
);
2116 in_pack
->array
[in_pack
->nr
].object
= object
;
2121 * Compare the objects in the offset order, in order to emulate the
2122 * "git rev-list --objects" output that produced the pack originally.
2124 static int ofscmp(const void *a_
, const void *b_
)
2126 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
2127 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
2129 if (a
->offset
< b
->offset
)
2131 else if (a
->offset
> b
->offset
)
2134 return hashcmp(a
->object
->sha1
, b
->object
->sha1
);
2137 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
2139 struct packed_git
*p
;
2140 struct in_pack in_pack
;
2143 memset(&in_pack
, 0, sizeof(in_pack
));
2145 for (p
= packed_git
; p
; p
= p
->next
) {
2146 const unsigned char *sha1
;
2149 if (!p
->pack_local
|| p
->pack_keep
)
2151 if (open_pack_index(p
))
2152 die("cannot open pack index");
2154 ALLOC_GROW(in_pack
.array
,
2155 in_pack
.nr
+ p
->num_objects
,
2158 for (i
= 0; i
< p
->num_objects
; i
++) {
2159 sha1
= nth_packed_object_sha1(p
, i
);
2160 o
= lookup_unknown_object(sha1
);
2161 if (!(o
->flags
& OBJECT_ADDED
))
2162 mark_in_pack_object(o
, p
, &in_pack
);
2163 o
->flags
|= OBJECT_ADDED
;
2168 qsort(in_pack
.array
, in_pack
.nr
, sizeof(in_pack
.array
[0]),
2170 for (i
= 0; i
< in_pack
.nr
; i
++) {
2171 struct object
*o
= in_pack
.array
[i
].object
;
2172 add_object_entry(o
->sha1
, o
->type
, "", 0);
2175 free(in_pack
.array
);
2178 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1
)
2180 static struct packed_git
*last_found
= (void *)1;
2181 struct packed_git
*p
;
2183 p
= (last_found
!= (void *)1) ? last_found
: packed_git
;
2186 if ((!p
->pack_local
|| p
->pack_keep
) &&
2187 find_pack_entry_one(sha1
, p
)) {
2191 if (p
== last_found
)
2195 if (p
== last_found
)
2201 static void loosen_unused_packed_objects(struct rev_info
*revs
)
2203 struct packed_git
*p
;
2205 const unsigned char *sha1
;
2207 for (p
= packed_git
; p
; p
= p
->next
) {
2208 if (!p
->pack_local
|| p
->pack_keep
)
2211 if (open_pack_index(p
))
2212 die("cannot open pack index");
2214 for (i
= 0; i
< p
->num_objects
; i
++) {
2215 sha1
= nth_packed_object_sha1(p
, i
);
2216 if (!locate_object_entry(sha1
) &&
2217 !has_sha1_pack_kept_or_nonlocal(sha1
))
2218 if (force_object_loose(sha1
, p
->mtime
))
2219 die("unable to force loose object");
2224 static void get_object_list(int ac
, const char **av
)
2226 struct rev_info revs
;
2230 init_revisions(&revs
, NULL
);
2231 save_commit_buffer
= 0;
2232 setup_revisions(ac
, av
, &revs
, NULL
);
2234 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2235 int len
= strlen(line
);
2236 if (len
&& line
[len
- 1] == '\n')
2241 if (!strcmp(line
, "--not")) {
2242 flags
^= UNINTERESTING
;
2245 die("not a rev '%s'", line
);
2247 if (handle_revision_arg(line
, &revs
, flags
, 1))
2248 die("bad revision '%s'", line
);
2251 if (prepare_revision_walk(&revs
))
2252 die("revision walk setup failed");
2253 mark_edges_uninteresting(revs
.commits
, &revs
, show_edge
);
2254 traverse_commit_list(&revs
, show_commit
, show_object
, NULL
);
2256 if (keep_unreachable
)
2257 add_objects_in_unpacked_packs(&revs
);
2258 if (unpack_unreachable
)
2259 loosen_unused_packed_objects(&revs
);
2262 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
2264 int use_internal_rev_list
= 0;
2266 int all_progress_implied
= 0;
2269 int rp_ac_alloc
= 64;
2272 read_replace_refs
= 0;
2274 rp_av
= xcalloc(rp_ac_alloc
, sizeof(*rp_av
));
2276 rp_av
[0] = "pack-objects";
2277 rp_av
[1] = "--objects"; /* --thin will make it --objects-edge */
2280 reset_pack_idx_option(&pack_idx_opts
);
2281 git_config(git_pack_config
, NULL
);
2282 if (!pack_compression_seen
&& core_compression_seen
)
2283 pack_compression_level
= core_compression_level
;
2285 progress
= isatty(2);
2286 for (i
= 1; i
< argc
; i
++) {
2287 const char *arg
= argv
[i
];
2292 if (!strcmp("--non-empty", arg
)) {
2296 if (!strcmp("--local", arg
)) {
2300 if (!strcmp("--incremental", arg
)) {
2304 if (!strcmp("--honor-pack-keep", arg
)) {
2305 ignore_packed_keep
= 1;
2308 if (!prefixcmp(arg
, "--compression=")) {
2310 int level
= strtoul(arg
+14, &end
, 0);
2311 if (!arg
[14] || *end
)
2314 level
= Z_DEFAULT_COMPRESSION
;
2315 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
2316 die("bad pack compression level %d", level
);
2317 pack_compression_level
= level
;
2320 if (!prefixcmp(arg
, "--max-pack-size=")) {
2321 pack_size_limit_cfg
= 0;
2322 if (!git_parse_ulong(arg
+16, &pack_size_limit
))
2326 if (!prefixcmp(arg
, "--window=")) {
2328 window
= strtoul(arg
+9, &end
, 0);
2329 if (!arg
[9] || *end
)
2333 if (!prefixcmp(arg
, "--window-memory=")) {
2334 if (!git_parse_ulong(arg
+16, &window_memory_limit
))
2338 if (!prefixcmp(arg
, "--threads=")) {
2340 delta_search_threads
= strtoul(arg
+10, &end
, 0);
2341 if (!arg
[10] || *end
|| delta_search_threads
< 0)
2344 if (delta_search_threads
!= 1)
2345 warning("no threads support, "
2346 "ignoring %s", arg
);
2350 if (!prefixcmp(arg
, "--depth=")) {
2352 depth
= strtoul(arg
+8, &end
, 0);
2353 if (!arg
[8] || *end
)
2357 if (!strcmp("--progress", arg
)) {
2361 if (!strcmp("--all-progress", arg
)) {
2365 if (!strcmp("--all-progress-implied", arg
)) {
2366 all_progress_implied
= 1;
2369 if (!strcmp("-q", arg
)) {
2373 if (!strcmp("--no-reuse-delta", arg
)) {
2377 if (!strcmp("--no-reuse-object", arg
)) {
2378 reuse_object
= reuse_delta
= 0;
2381 if (!strcmp("--delta-base-offset", arg
)) {
2382 allow_ofs_delta
= 1;
2385 if (!strcmp("--stdout", arg
)) {
2389 if (!strcmp("--revs", arg
)) {
2390 use_internal_rev_list
= 1;
2393 if (!strcmp("--keep-unreachable", arg
)) {
2394 keep_unreachable
= 1;
2397 if (!strcmp("--unpack-unreachable", arg
)) {
2398 unpack_unreachable
= 1;
2401 if (!strcmp("--include-tag", arg
)) {
2405 if (!strcmp("--unpacked", arg
) ||
2406 !strcmp("--reflog", arg
) ||
2407 !strcmp("--all", arg
)) {
2408 use_internal_rev_list
= 1;
2409 if (rp_ac
>= rp_ac_alloc
- 1) {
2410 rp_ac_alloc
= alloc_nr(rp_ac_alloc
);
2411 rp_av
= xrealloc(rp_av
,
2412 rp_ac_alloc
* sizeof(*rp_av
));
2414 rp_av
[rp_ac
++] = arg
;
2417 if (!strcmp("--thin", arg
)) {
2418 use_internal_rev_list
= 1;
2420 rp_av
[1] = "--objects-edge";
2423 if (!prefixcmp(arg
, "--index-version=")) {
2425 pack_idx_opts
.version
= strtoul(arg
+ 16, &c
, 10);
2426 if (pack_idx_opts
.version
> 2)
2429 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
2430 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
2434 if (!strcmp(arg
, "--keep-true-parents")) {
2435 grafts_replace_parents
= 0;
2441 /* Traditionally "pack-objects [options] base extra" failed;
2442 * we would however want to take refs parameter that would
2443 * have been given to upstream rev-list ourselves, which means
2444 * we somehow want to say what the base name is. So the
2447 * pack-objects [options] base <refs...>
2449 * in other words, we would treat the first non-option as the
2450 * base_name and send everything else to the internal revision
2454 if (!pack_to_stdout
)
2455 base_name
= argv
[i
++];
2457 if (pack_to_stdout
!= !base_name
)
2460 if (!pack_to_stdout
&& !pack_size_limit
)
2461 pack_size_limit
= pack_size_limit_cfg
;
2462 if (pack_to_stdout
&& pack_size_limit
)
2463 die("--max-pack-size cannot be used to build a pack for transfer.");
2464 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
2465 warning("minimum pack size limit is 1 MiB");
2466 pack_size_limit
= 1024*1024;
2469 if (!pack_to_stdout
&& thin
)
2470 die("--thin cannot be used to build an indexable pack.");
2472 if (keep_unreachable
&& unpack_unreachable
)
2473 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2475 if (progress
&& all_progress_implied
)
2478 prepare_packed_git();
2481 progress_state
= start_progress("Counting objects", 0);
2482 if (!use_internal_rev_list
)
2483 read_object_list_from_stdin();
2485 rp_av
[rp_ac
] = NULL
;
2486 get_object_list(rp_ac
, rp_av
);
2488 cleanup_preferred_base();
2489 if (include_tag
&& nr_result
)
2490 for_each_ref(add_ref_tag
, NULL
);
2491 stop_progress(&progress_state
);
2493 if (non_empty
&& !nr_result
)
2496 prepare_pack(window
, depth
);
2499 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
2500 " reused %"PRIu32
" (delta %"PRIu32
")\n",
2501 written
, written_delta
, reused
, reused_delta
);