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 --stdout [options...] [< ref-list | < object-list]",
23 "git pack-objects [options...] base-name [< ref-list | < object-list]",
28 struct pack_idx_entry idx
;
29 unsigned long size
; /* uncompressed size */
30 struct packed_git
*in_pack
; /* already in pack */
32 struct object_entry
*delta
; /* delta base object */
33 struct object_entry
*delta_child
; /* deltified objects who bases me */
34 struct object_entry
*delta_sibling
; /* other deltified objects who
35 * uses the same base as me
37 void *delta_data
; /* cached delta (uncompressed) */
38 unsigned long delta_size
; /* delta data size (uncompressed) */
39 unsigned long z_delta_size
; /* delta data size (compressed) */
40 unsigned int hash
; /* name hint hash */
41 enum object_type type
;
42 enum object_type in_pack_type
; /* could be delta */
43 unsigned char in_pack_header_size
;
44 unsigned char preferred_base
; /* we do not pack this, but is available
45 * to be used as the base object to delta
48 unsigned char no_try_delta
;
49 unsigned char tagged
; /* near the very tip of refs */
50 unsigned char filled
; /* assigned write-order */
54 * Objects we are going to pack are collected in objects array (dynamically
55 * expanded). nr_objects & nr_alloc controls this array. They are stored
56 * in the order we see -- typically rev-list --objects order that gives us
57 * nice "minimum seek" order.
59 static struct object_entry
*objects
;
60 static struct pack_idx_entry
**written_list
;
61 static uint32_t nr_objects
, nr_alloc
, nr_result
, nr_written
;
64 static int reuse_delta
= 1, reuse_object
= 1;
65 static int keep_unreachable
, unpack_unreachable
, include_tag
;
66 static unsigned long unpack_unreachable_expiration
;
68 static int incremental
;
69 static int ignore_packed_keep
;
70 static int allow_ofs_delta
;
71 static struct pack_idx_option pack_idx_opts
;
72 static const char *base_name
;
73 static int progress
= 1;
74 static int window
= 10;
75 static unsigned long pack_size_limit
;
76 static int depth
= 50;
77 static int delta_search_threads
;
78 static int pack_to_stdout
;
79 static int num_preferred_base
;
80 static struct progress
*progress_state
;
81 static int pack_compression_level
= Z_DEFAULT_COMPRESSION
;
82 static int pack_compression_seen
;
84 static unsigned long delta_cache_size
= 0;
85 static unsigned long max_delta_cache_size
= 256 * 1024 * 1024;
86 static unsigned long cache_max_small_delta_size
= 1000;
88 static unsigned long window_memory_limit
= 0;
91 * The object names in objects array are hashed with this hashtable,
92 * to help looking up the entry by object name.
93 * This hashtable is built after all the objects are seen.
95 static int *object_ix
;
96 static int object_ix_hashsz
;
97 static struct object_entry
*locate_object_entry(const unsigned char *sha1
);
102 static uint32_t written
, written_delta
;
103 static uint32_t reused
, reused_delta
;
106 static void *get_delta(struct object_entry
*entry
)
108 unsigned long size
, base_size
, delta_size
;
109 void *buf
, *base_buf
, *delta_buf
;
110 enum object_type type
;
112 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
114 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
115 base_buf
= read_sha1_file(entry
->delta
->idx
.sha1
, &type
, &base_size
);
117 die("unable to read %s", sha1_to_hex(entry
->delta
->idx
.sha1
));
118 delta_buf
= diff_delta(base_buf
, base_size
,
119 buf
, size
, &delta_size
, 0);
120 if (!delta_buf
|| delta_size
!= entry
->delta_size
)
121 die("delta size changed");
127 static unsigned long do_compress(void **pptr
, unsigned long size
)
131 unsigned long maxsize
;
133 memset(&stream
, 0, sizeof(stream
));
134 git_deflate_init(&stream
, pack_compression_level
);
135 maxsize
= git_deflate_bound(&stream
, size
);
138 out
= xmalloc(maxsize
);
142 stream
.avail_in
= size
;
143 stream
.next_out
= out
;
144 stream
.avail_out
= maxsize
;
145 while (git_deflate(&stream
, Z_FINISH
) == Z_OK
)
147 git_deflate_end(&stream
);
150 return stream
.total_out
;
154 * we are going to reuse the existing object data as is. make
155 * sure it is not corrupt.
157 static int check_pack_inflate(struct packed_git
*p
,
158 struct pack_window
**w_curs
,
161 unsigned long expect
)
164 unsigned char fakebuf
[4096], *in
;
167 memset(&stream
, 0, sizeof(stream
));
168 git_inflate_init(&stream
);
170 in
= use_pack(p
, w_curs
, offset
, &stream
.avail_in
);
172 stream
.next_out
= fakebuf
;
173 stream
.avail_out
= sizeof(fakebuf
);
174 st
= git_inflate(&stream
, Z_FINISH
);
175 offset
+= stream
.next_in
- in
;
176 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
177 git_inflate_end(&stream
);
178 return (st
== Z_STREAM_END
&&
179 stream
.total_out
== expect
&&
180 stream
.total_in
== len
) ? 0 : -1;
183 static void copy_pack_data(struct sha1file
*f
,
184 struct packed_git
*p
,
185 struct pack_window
**w_curs
,
193 in
= use_pack(p
, w_curs
, offset
, &avail
);
195 avail
= (unsigned long)len
;
196 sha1write(f
, in
, avail
);
202 /* Return 0 if we will bust the pack-size limit */
203 static unsigned long write_no_reuse_object(struct sha1file
*f
, struct object_entry
*entry
,
204 unsigned long limit
, int usable_delta
)
206 unsigned long size
, datalen
;
207 unsigned char header
[10], dheader
[10];
209 enum object_type type
;
213 buf
= read_sha1_file(entry
->idx
.sha1
, &type
, &size
);
215 die("unable to read %s", sha1_to_hex(entry
->idx
.sha1
));
217 * make sure no cached delta data remains from a
218 * previous attempt before a pack split occurred.
220 free(entry
->delta_data
);
221 entry
->delta_data
= NULL
;
222 entry
->z_delta_size
= 0;
223 } else if (entry
->delta_data
) {
224 size
= entry
->delta_size
;
225 buf
= entry
->delta_data
;
226 entry
->delta_data
= NULL
;
227 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
228 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
230 buf
= get_delta(entry
);
231 size
= entry
->delta_size
;
232 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
233 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
236 if (entry
->z_delta_size
)
237 datalen
= entry
->z_delta_size
;
239 datalen
= do_compress(&buf
, size
);
242 * The object header is a byte of 'type' followed by zero or
243 * more bytes of length.
245 hdrlen
= encode_in_pack_object_header(type
, size
, header
);
247 if (type
== OBJ_OFS_DELTA
) {
249 * Deltas with relative base contain an additional
250 * encoding of the relative offset for the delta
251 * base from this object's position in the pack.
253 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
254 unsigned pos
= sizeof(dheader
) - 1;
255 dheader
[pos
] = ofs
& 127;
257 dheader
[--pos
] = 128 | (--ofs
& 127);
258 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
262 sha1write(f
, header
, hdrlen
);
263 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
264 hdrlen
+= sizeof(dheader
) - pos
;
265 } else if (type
== OBJ_REF_DELTA
) {
267 * Deltas with a base reference contain
268 * an additional 20 bytes for the base sha1.
270 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
274 sha1write(f
, header
, hdrlen
);
275 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
278 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
282 sha1write(f
, header
, hdrlen
);
284 sha1write(f
, buf
, datalen
);
287 return hdrlen
+ datalen
;
290 /* Return 0 if we will bust the pack-size limit */
291 static unsigned long write_reuse_object(struct sha1file
*f
, struct object_entry
*entry
,
292 unsigned long limit
, int usable_delta
)
294 struct packed_git
*p
= entry
->in_pack
;
295 struct pack_window
*w_curs
= NULL
;
296 struct revindex_entry
*revidx
;
298 enum object_type type
= entry
->type
;
299 unsigned long datalen
;
300 unsigned char header
[10], dheader
[10];
304 type
= (allow_ofs_delta
&& entry
->delta
->idx
.offset
) ?
305 OBJ_OFS_DELTA
: OBJ_REF_DELTA
;
306 hdrlen
= encode_in_pack_object_header(type
, entry
->size
, header
);
308 offset
= entry
->in_pack_offset
;
309 revidx
= find_pack_revindex(p
, offset
);
310 datalen
= revidx
[1].offset
- offset
;
311 if (!pack_to_stdout
&& p
->index_version
> 1 &&
312 check_pack_crc(p
, &w_curs
, offset
, datalen
, revidx
->nr
)) {
313 error("bad packed object CRC for %s", sha1_to_hex(entry
->idx
.sha1
));
315 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
318 offset
+= entry
->in_pack_header_size
;
319 datalen
-= entry
->in_pack_header_size
;
321 if (!pack_to_stdout
&& p
->index_version
== 1 &&
322 check_pack_inflate(p
, &w_curs
, offset
, datalen
, entry
->size
)) {
323 error("corrupt packed object for %s", sha1_to_hex(entry
->idx
.sha1
));
325 return write_no_reuse_object(f
, entry
, limit
, usable_delta
);
328 if (type
== OBJ_OFS_DELTA
) {
329 off_t ofs
= entry
->idx
.offset
- entry
->delta
->idx
.offset
;
330 unsigned pos
= sizeof(dheader
) - 1;
331 dheader
[pos
] = ofs
& 127;
333 dheader
[--pos
] = 128 | (--ofs
& 127);
334 if (limit
&& hdrlen
+ sizeof(dheader
) - pos
+ datalen
+ 20 >= limit
) {
338 sha1write(f
, header
, hdrlen
);
339 sha1write(f
, dheader
+ pos
, sizeof(dheader
) - pos
);
340 hdrlen
+= sizeof(dheader
) - pos
;
342 } else if (type
== OBJ_REF_DELTA
) {
343 if (limit
&& hdrlen
+ 20 + datalen
+ 20 >= limit
) {
347 sha1write(f
, header
, hdrlen
);
348 sha1write(f
, entry
->delta
->idx
.sha1
, 20);
352 if (limit
&& hdrlen
+ datalen
+ 20 >= limit
) {
356 sha1write(f
, header
, hdrlen
);
358 copy_pack_data(f
, p
, &w_curs
, offset
, datalen
);
361 return hdrlen
+ datalen
;
364 /* Return 0 if we will bust the pack-size limit */
365 static unsigned long write_object(struct sha1file
*f
,
366 struct object_entry
*entry
,
369 unsigned long limit
, len
;
370 int usable_delta
, to_reuse
;
375 /* apply size limit if limited packsize and not first object */
376 if (!pack_size_limit
|| !nr_written
)
378 else if (pack_size_limit
<= write_offset
)
380 * the earlier object did not fit the limit; avoid
381 * mistaking this with unlimited (i.e. limit = 0).
385 limit
= pack_size_limit
- write_offset
;
388 usable_delta
= 0; /* no delta */
389 else if (!pack_size_limit
)
390 usable_delta
= 1; /* unlimited packfile */
391 else if (entry
->delta
->idx
.offset
== (off_t
)-1)
392 usable_delta
= 0; /* base was written to another pack */
393 else if (entry
->delta
->idx
.offset
)
394 usable_delta
= 1; /* base already exists in this pack */
396 usable_delta
= 0; /* base could end up in another pack */
399 to_reuse
= 0; /* explicit */
400 else if (!entry
->in_pack
)
401 to_reuse
= 0; /* can't reuse what we don't have */
402 else if (entry
->type
== OBJ_REF_DELTA
|| entry
->type
== OBJ_OFS_DELTA
)
403 /* check_object() decided it for us ... */
404 to_reuse
= usable_delta
;
405 /* ... but pack split may override that */
406 else if (entry
->type
!= entry
->in_pack_type
)
407 to_reuse
= 0; /* pack has delta which is unusable */
408 else if (entry
->delta
)
409 to_reuse
= 0; /* we want to pack afresh */
411 to_reuse
= 1; /* we have it in-pack undeltified,
412 * and we do not need to deltify it.
416 len
= write_no_reuse_object(f
, entry
, limit
, usable_delta
);
418 len
= write_reuse_object(f
, entry
, limit
, usable_delta
);
426 entry
->idx
.crc32
= crc32_end(f
);
430 enum write_one_status
{
431 WRITE_ONE_SKIP
= -1, /* already written */
432 WRITE_ONE_BREAK
= 0, /* writing this will bust the limit; not written */
433 WRITE_ONE_WRITTEN
= 1, /* normal */
434 WRITE_ONE_RECURSIVE
= 2 /* already scheduled to be written */
437 static enum write_one_status
write_one(struct sha1file
*f
,
438 struct object_entry
*e
,
445 * we set offset to 1 (which is an impossible value) to mark
446 * the fact that this object is involved in "write its base
447 * first before writing a deltified object" recursion.
449 recursing
= (e
->idx
.offset
== 1);
451 warning("recursive delta detected for object %s",
452 sha1_to_hex(e
->idx
.sha1
));
453 return WRITE_ONE_RECURSIVE
;
454 } else if (e
->idx
.offset
|| e
->preferred_base
) {
455 /* offset is non zero if object is written already. */
456 return WRITE_ONE_SKIP
;
459 /* if we are deltified, write out base object first. */
461 e
->idx
.offset
= 1; /* now recurse */
462 switch (write_one(f
, e
->delta
, offset
)) {
463 case WRITE_ONE_RECURSIVE
:
464 /* we cannot depend on this one */
469 case WRITE_ONE_BREAK
:
470 e
->idx
.offset
= recursing
;
471 return WRITE_ONE_BREAK
;
475 e
->idx
.offset
= *offset
;
476 size
= write_object(f
, e
, *offset
);
478 e
->idx
.offset
= recursing
;
479 return WRITE_ONE_BREAK
;
481 written_list
[nr_written
++] = &e
->idx
;
483 /* make sure off_t is sufficiently large not to wrap */
484 if (signed_add_overflows(*offset
, size
))
485 die("pack too large for current definition of off_t");
487 return WRITE_ONE_WRITTEN
;
490 static int mark_tagged(const char *path
, const unsigned char *sha1
, int flag
,
493 unsigned char peeled
[20];
494 struct object_entry
*entry
= locate_object_entry(sha1
);
498 if (!peel_ref(path
, peeled
)) {
499 entry
= locate_object_entry(peeled
);
506 static inline void add_to_write_order(struct object_entry
**wo
,
508 struct object_entry
*e
)
516 static void add_descendants_to_write_order(struct object_entry
**wo
,
518 struct object_entry
*e
)
520 int add_to_order
= 1;
523 struct object_entry
*s
;
524 /* add this node... */
525 add_to_write_order(wo
, endp
, e
);
526 /* all its siblings... */
527 for (s
= e
->delta_sibling
; s
; s
= s
->delta_sibling
) {
528 add_to_write_order(wo
, endp
, s
);
531 /* drop down a level to add left subtree nodes if possible */
532 if (e
->delta_child
) {
537 /* our sibling might have some children, it is next */
538 if (e
->delta_sibling
) {
539 e
= e
->delta_sibling
;
542 /* go back to our parent node */
544 while (e
&& !e
->delta_sibling
) {
545 /* we're on the right side of a subtree, keep
546 * going up until we can go right again */
550 /* done- we hit our original root node */
553 /* pass it off to sibling at this level */
554 e
= e
->delta_sibling
;
559 static void add_family_to_write_order(struct object_entry
**wo
,
561 struct object_entry
*e
)
563 struct object_entry
*root
;
565 for (root
= e
; root
->delta
; root
= root
->delta
)
567 add_descendants_to_write_order(wo
, endp
, root
);
570 static struct object_entry
**compute_write_order(void)
572 unsigned int i
, wo_end
, last_untagged
;
574 struct object_entry
**wo
= xmalloc(nr_objects
* sizeof(*wo
));
576 for (i
= 0; i
< nr_objects
; i
++) {
577 objects
[i
].tagged
= 0;
578 objects
[i
].filled
= 0;
579 objects
[i
].delta_child
= NULL
;
580 objects
[i
].delta_sibling
= NULL
;
584 * Fully connect delta_child/delta_sibling network.
585 * Make sure delta_sibling is sorted in the original
588 for (i
= nr_objects
; i
> 0;) {
589 struct object_entry
*e
= &objects
[--i
];
592 /* Mark me as the first child */
593 e
->delta_sibling
= e
->delta
->delta_child
;
594 e
->delta
->delta_child
= e
;
598 * Mark objects that are at the tip of tags.
600 for_each_tag_ref(mark_tagged
, NULL
);
603 * Give the objects in the original recency order until
604 * we see a tagged tip.
606 for (i
= wo_end
= 0; i
< nr_objects
; i
++) {
607 if (objects
[i
].tagged
)
609 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
614 * Then fill all the tagged tips.
616 for (; i
< nr_objects
; i
++) {
617 if (objects
[i
].tagged
)
618 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
622 * And then all remaining commits and tags.
624 for (i
= last_untagged
; i
< nr_objects
; i
++) {
625 if (objects
[i
].type
!= OBJ_COMMIT
&&
626 objects
[i
].type
!= OBJ_TAG
)
628 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
632 * And then all the trees.
634 for (i
= last_untagged
; i
< nr_objects
; i
++) {
635 if (objects
[i
].type
!= OBJ_TREE
)
637 add_to_write_order(wo
, &wo_end
, &objects
[i
]);
641 * Finally all the rest in really tight order
643 for (i
= last_untagged
; i
< nr_objects
; i
++) {
644 if (!objects
[i
].filled
)
645 add_family_to_write_order(wo
, &wo_end
, &objects
[i
]);
648 if (wo_end
!= nr_objects
)
649 die("ordered %u objects, expected %"PRIu32
, wo_end
, nr_objects
);
654 static void write_pack_file(void)
659 uint32_t nr_remaining
= nr_result
;
660 time_t last_mtime
= 0;
661 struct object_entry
**write_order
;
663 if (progress
> pack_to_stdout
)
664 progress_state
= start_progress("Writing objects", nr_result
);
665 written_list
= xmalloc(nr_objects
* sizeof(*written_list
));
666 write_order
= compute_write_order();
669 unsigned char sha1
[20];
670 char *pack_tmp_name
= NULL
;
673 f
= sha1fd_throughput(1, "<stdout>", progress_state
);
675 f
= create_tmp_packfile(&pack_tmp_name
);
677 offset
= write_pack_header(f
, nr_remaining
);
679 die_errno("unable to write pack header");
681 for (; i
< nr_objects
; i
++) {
682 struct object_entry
*e
= write_order
[i
];
683 if (write_one(f
, e
, &offset
) == WRITE_ONE_BREAK
)
685 display_progress(progress_state
, written
);
689 * Did we write the wrong # entries in the header?
690 * If so, rewrite it like in fast-import
692 if (pack_to_stdout
) {
693 sha1close(f
, sha1
, CSUM_CLOSE
);
694 } else if (nr_written
== nr_remaining
) {
695 sha1close(f
, sha1
, CSUM_FSYNC
);
697 int fd
= sha1close(f
, sha1
, 0);
698 fixup_pack_header_footer(fd
, sha1
, pack_tmp_name
,
699 nr_written
, sha1
, offset
);
703 if (!pack_to_stdout
) {
705 char tmpname
[PATH_MAX
];
708 * Packs are runtime accessed in their mtime
709 * order since newer packs are more likely to contain
710 * younger objects. So if we are creating multiple
711 * packs then we should modify the mtime of later ones
712 * to preserve this property.
714 if (stat(pack_tmp_name
, &st
) < 0) {
715 warning("failed to stat %s: %s",
716 pack_tmp_name
, strerror(errno
));
717 } else if (!last_mtime
) {
718 last_mtime
= st
.st_mtime
;
721 utb
.actime
= st
.st_atime
;
722 utb
.modtime
= --last_mtime
;
723 if (utime(pack_tmp_name
, &utb
) < 0)
724 warning("failed utime() on %s: %s",
725 tmpname
, strerror(errno
));
728 /* Enough space for "-<sha-1>.pack"? */
729 if (sizeof(tmpname
) <= strlen(base_name
) + 50)
730 die("pack base name '%s' too long", base_name
);
731 snprintf(tmpname
, sizeof(tmpname
), "%s-", base_name
);
732 finish_tmp_packfile(tmpname
, pack_tmp_name
,
733 written_list
, nr_written
,
734 &pack_idx_opts
, sha1
);
736 puts(sha1_to_hex(sha1
));
739 /* mark written objects as written to previous pack */
740 for (j
= 0; j
< nr_written
; j
++) {
741 written_list
[j
]->offset
= (off_t
)-1;
743 nr_remaining
-= nr_written
;
744 } while (nr_remaining
&& i
< nr_objects
);
748 stop_progress(&progress_state
);
749 if (written
!= nr_result
)
750 die("wrote %"PRIu32
" objects while expecting %"PRIu32
,
754 static int locate_object_entry_hash(const unsigned char *sha1
)
758 memcpy(&ui
, sha1
, sizeof(unsigned int));
759 i
= ui
% object_ix_hashsz
;
760 while (0 < object_ix
[i
]) {
761 if (!hashcmp(sha1
, objects
[object_ix
[i
] - 1].idx
.sha1
))
763 if (++i
== object_ix_hashsz
)
769 static struct object_entry
*locate_object_entry(const unsigned char *sha1
)
773 if (!object_ix_hashsz
)
776 i
= locate_object_entry_hash(sha1
);
778 return &objects
[object_ix
[i
]-1];
782 static void rehash_objects(void)
785 struct object_entry
*oe
;
787 object_ix_hashsz
= nr_objects
* 3;
788 if (object_ix_hashsz
< 1024)
789 object_ix_hashsz
= 1024;
790 object_ix
= xrealloc(object_ix
, sizeof(int) * object_ix_hashsz
);
791 memset(object_ix
, 0, sizeof(int) * object_ix_hashsz
);
792 for (i
= 0, oe
= objects
; i
< nr_objects
; i
++, oe
++) {
793 int ix
= locate_object_entry_hash(oe
->idx
.sha1
);
797 object_ix
[ix
] = i
+ 1;
801 static unsigned name_hash(const char *name
)
803 unsigned c
, hash
= 0;
809 * This effectively just creates a sortable number from the
810 * last sixteen non-whitespace characters. Last characters
811 * count "most", so things that end in ".c" sort together.
813 while ((c
= *name
++) != 0) {
816 hash
= (hash
>> 2) + (c
<< 24);
821 static void setup_delta_attr_check(struct git_attr_check
*check
)
823 static struct git_attr
*attr_delta
;
826 attr_delta
= git_attr("delta");
828 check
[0].attr
= attr_delta
;
831 static int no_try_delta(const char *path
)
833 struct git_attr_check check
[1];
835 setup_delta_attr_check(check
);
836 if (git_check_attr(path
, ARRAY_SIZE(check
), check
))
838 if (ATTR_FALSE(check
->value
))
843 static int add_object_entry(const unsigned char *sha1
, enum object_type type
,
844 const char *name
, int exclude
)
846 struct object_entry
*entry
;
847 struct packed_git
*p
, *found_pack
= NULL
;
848 off_t found_offset
= 0;
850 unsigned hash
= name_hash(name
);
852 ix
= nr_objects
? locate_object_entry_hash(sha1
) : -1;
855 entry
= objects
+ object_ix
[ix
] - 1;
856 if (!entry
->preferred_base
)
858 entry
->preferred_base
= 1;
863 if (!exclude
&& local
&& has_loose_object_nonlocal(sha1
))
866 for (p
= packed_git
; p
; p
= p
->next
) {
867 off_t offset
= find_pack_entry_one(sha1
, p
);
870 if (!is_pack_valid(p
)) {
871 warning("packfile %s cannot be accessed", p
->pack_name
);
874 found_offset
= offset
;
881 if (local
&& !p
->pack_local
)
883 if (ignore_packed_keep
&& p
->pack_local
&& p
->pack_keep
)
888 if (nr_objects
>= nr_alloc
) {
889 nr_alloc
= (nr_alloc
+ 1024) * 3 / 2;
890 objects
= xrealloc(objects
, nr_alloc
* sizeof(*entry
));
893 entry
= objects
+ nr_objects
++;
894 memset(entry
, 0, sizeof(*entry
));
895 hashcpy(entry
->idx
.sha1
, sha1
);
900 entry
->preferred_base
= 1;
904 entry
->in_pack
= found_pack
;
905 entry
->in_pack_offset
= found_offset
;
908 if (object_ix_hashsz
* 3 <= nr_objects
* 4)
911 object_ix
[-1 - ix
] = nr_objects
;
913 display_progress(progress_state
, nr_objects
);
915 if (name
&& no_try_delta(name
))
916 entry
->no_try_delta
= 1;
921 struct pbase_tree_cache
{
922 unsigned char sha1
[20];
926 unsigned long tree_size
;
929 static struct pbase_tree_cache
*(pbase_tree_cache
[256]);
930 static int pbase_tree_cache_ix(const unsigned char *sha1
)
932 return sha1
[0] % ARRAY_SIZE(pbase_tree_cache
);
934 static int pbase_tree_cache_ix_incr(int ix
)
936 return (ix
+1) % ARRAY_SIZE(pbase_tree_cache
);
939 static struct pbase_tree
{
940 struct pbase_tree
*next
;
941 /* This is a phony "cache" entry; we are not
942 * going to evict it nor find it through _get()
943 * mechanism -- this is for the toplevel node that
944 * would almost always change with any commit.
946 struct pbase_tree_cache pcache
;
949 static struct pbase_tree_cache
*pbase_tree_get(const unsigned char *sha1
)
951 struct pbase_tree_cache
*ent
, *nent
;
954 enum object_type type
;
956 int my_ix
= pbase_tree_cache_ix(sha1
);
957 int available_ix
= -1;
959 /* pbase-tree-cache acts as a limited hashtable.
960 * your object will be found at your index or within a few
961 * slots after that slot if it is cached.
963 for (neigh
= 0; neigh
< 8; neigh
++) {
964 ent
= pbase_tree_cache
[my_ix
];
965 if (ent
&& !hashcmp(ent
->sha1
, sha1
)) {
969 else if (((available_ix
< 0) && (!ent
|| !ent
->ref
)) ||
970 ((0 <= available_ix
) &&
971 (!ent
&& pbase_tree_cache
[available_ix
])))
972 available_ix
= my_ix
;
975 my_ix
= pbase_tree_cache_ix_incr(my_ix
);
978 /* Did not find one. Either we got a bogus request or
979 * we need to read and perhaps cache.
981 data
= read_sha1_file(sha1
, &type
, &size
);
984 if (type
!= OBJ_TREE
) {
989 /* We need to either cache or return a throwaway copy */
991 if (available_ix
< 0)
994 ent
= pbase_tree_cache
[available_ix
];
995 my_ix
= available_ix
;
999 nent
= xmalloc(sizeof(*nent
));
1000 nent
->temporary
= (available_ix
< 0);
1003 /* evict and reuse */
1004 free(ent
->tree_data
);
1007 hashcpy(nent
->sha1
, sha1
);
1008 nent
->tree_data
= data
;
1009 nent
->tree_size
= size
;
1011 if (!nent
->temporary
)
1012 pbase_tree_cache
[my_ix
] = nent
;
1016 static void pbase_tree_put(struct pbase_tree_cache
*cache
)
1018 if (!cache
->temporary
) {
1022 free(cache
->tree_data
);
1026 static int name_cmp_len(const char *name
)
1029 for (i
= 0; name
[i
] && name
[i
] != '\n' && name
[i
] != '/'; i
++)
1034 static void add_pbase_object(struct tree_desc
*tree
,
1037 const char *fullname
)
1039 struct name_entry entry
;
1042 while (tree_entry(tree
,&entry
)) {
1043 if (S_ISGITLINK(entry
.mode
))
1045 cmp
= tree_entry_len(&entry
) != cmplen
? 1 :
1046 memcmp(name
, entry
.path
, cmplen
);
1051 if (name
[cmplen
] != '/') {
1052 add_object_entry(entry
.sha1
,
1053 object_type(entry
.mode
),
1057 if (S_ISDIR(entry
.mode
)) {
1058 struct tree_desc sub
;
1059 struct pbase_tree_cache
*tree
;
1060 const char *down
= name
+cmplen
+1;
1061 int downlen
= name_cmp_len(down
);
1063 tree
= pbase_tree_get(entry
.sha1
);
1066 init_tree_desc(&sub
, tree
->tree_data
, tree
->tree_size
);
1068 add_pbase_object(&sub
, down
, downlen
, fullname
);
1069 pbase_tree_put(tree
);
1074 static unsigned *done_pbase_paths
;
1075 static int done_pbase_paths_num
;
1076 static int done_pbase_paths_alloc
;
1077 static int done_pbase_path_pos(unsigned hash
)
1080 int hi
= done_pbase_paths_num
;
1082 int mi
= (hi
+ lo
) / 2;
1083 if (done_pbase_paths
[mi
] == hash
)
1085 if (done_pbase_paths
[mi
] < hash
)
1093 static int check_pbase_path(unsigned hash
)
1095 int pos
= (!done_pbase_paths
) ? -1 : done_pbase_path_pos(hash
);
1099 if (done_pbase_paths_alloc
<= done_pbase_paths_num
) {
1100 done_pbase_paths_alloc
= alloc_nr(done_pbase_paths_alloc
);
1101 done_pbase_paths
= xrealloc(done_pbase_paths
,
1102 done_pbase_paths_alloc
*
1105 done_pbase_paths_num
++;
1106 if (pos
< done_pbase_paths_num
)
1107 memmove(done_pbase_paths
+ pos
+ 1,
1108 done_pbase_paths
+ pos
,
1109 (done_pbase_paths_num
- pos
- 1) * sizeof(unsigned));
1110 done_pbase_paths
[pos
] = hash
;
1114 static void add_preferred_base_object(const char *name
)
1116 struct pbase_tree
*it
;
1118 unsigned hash
= name_hash(name
);
1120 if (!num_preferred_base
|| check_pbase_path(hash
))
1123 cmplen
= name_cmp_len(name
);
1124 for (it
= pbase_tree
; it
; it
= it
->next
) {
1126 add_object_entry(it
->pcache
.sha1
, OBJ_TREE
, NULL
, 1);
1129 struct tree_desc tree
;
1130 init_tree_desc(&tree
, it
->pcache
.tree_data
, it
->pcache
.tree_size
);
1131 add_pbase_object(&tree
, name
, cmplen
, name
);
1136 static void add_preferred_base(unsigned char *sha1
)
1138 struct pbase_tree
*it
;
1141 unsigned char tree_sha1
[20];
1143 if (window
<= num_preferred_base
++)
1146 data
= read_object_with_reference(sha1
, tree_type
, &size
, tree_sha1
);
1150 for (it
= pbase_tree
; it
; it
= it
->next
) {
1151 if (!hashcmp(it
->pcache
.sha1
, tree_sha1
)) {
1157 it
= xcalloc(1, sizeof(*it
));
1158 it
->next
= pbase_tree
;
1161 hashcpy(it
->pcache
.sha1
, tree_sha1
);
1162 it
->pcache
.tree_data
= data
;
1163 it
->pcache
.tree_size
= size
;
1166 static void cleanup_preferred_base(void)
1168 struct pbase_tree
*it
;
1174 struct pbase_tree
*this = it
;
1176 free(this->pcache
.tree_data
);
1180 for (i
= 0; i
< ARRAY_SIZE(pbase_tree_cache
); i
++) {
1181 if (!pbase_tree_cache
[i
])
1183 free(pbase_tree_cache
[i
]->tree_data
);
1184 free(pbase_tree_cache
[i
]);
1185 pbase_tree_cache
[i
] = NULL
;
1188 free(done_pbase_paths
);
1189 done_pbase_paths
= NULL
;
1190 done_pbase_paths_num
= done_pbase_paths_alloc
= 0;
1193 static void check_object(struct object_entry
*entry
)
1195 if (entry
->in_pack
) {
1196 struct packed_git
*p
= entry
->in_pack
;
1197 struct pack_window
*w_curs
= NULL
;
1198 const unsigned char *base_ref
= NULL
;
1199 struct object_entry
*base_entry
;
1200 unsigned long used
, used_0
;
1201 unsigned long avail
;
1203 unsigned char *buf
, c
;
1205 buf
= use_pack(p
, &w_curs
, entry
->in_pack_offset
, &avail
);
1208 * We want in_pack_type even if we do not reuse delta
1209 * since non-delta representations could still be reused.
1211 used
= unpack_object_header_buffer(buf
, avail
,
1212 &entry
->in_pack_type
,
1218 * Determine if this is a delta and if so whether we can
1219 * reuse it or not. Otherwise let's find out as cheaply as
1220 * possible what the actual type and size for this object is.
1222 switch (entry
->in_pack_type
) {
1224 /* Not a delta hence we've already got all we need. */
1225 entry
->type
= entry
->in_pack_type
;
1226 entry
->in_pack_header_size
= used
;
1227 if (entry
->type
< OBJ_COMMIT
|| entry
->type
> OBJ_BLOB
)
1229 unuse_pack(&w_curs
);
1232 if (reuse_delta
&& !entry
->preferred_base
)
1233 base_ref
= use_pack(p
, &w_curs
,
1234 entry
->in_pack_offset
+ used
, NULL
);
1235 entry
->in_pack_header_size
= used
+ 20;
1238 buf
= use_pack(p
, &w_curs
,
1239 entry
->in_pack_offset
+ used
, NULL
);
1245 if (!ofs
|| MSB(ofs
, 7)) {
1246 error("delta base offset overflow in pack for %s",
1247 sha1_to_hex(entry
->idx
.sha1
));
1251 ofs
= (ofs
<< 7) + (c
& 127);
1253 ofs
= entry
->in_pack_offset
- ofs
;
1254 if (ofs
<= 0 || ofs
>= entry
->in_pack_offset
) {
1255 error("delta base offset out of bound for %s",
1256 sha1_to_hex(entry
->idx
.sha1
));
1259 if (reuse_delta
&& !entry
->preferred_base
) {
1260 struct revindex_entry
*revidx
;
1261 revidx
= find_pack_revindex(p
, ofs
);
1264 base_ref
= nth_packed_object_sha1(p
, revidx
->nr
);
1266 entry
->in_pack_header_size
= used
+ used_0
;
1270 if (base_ref
&& (base_entry
= locate_object_entry(base_ref
))) {
1272 * If base_ref was set above that means we wish to
1273 * reuse delta data, and we even found that base
1274 * in the list of objects we want to pack. Goodie!
1276 * Depth value does not matter - find_deltas() will
1277 * never consider reused delta as the base object to
1278 * deltify other objects against, in order to avoid
1281 entry
->type
= entry
->in_pack_type
;
1282 entry
->delta
= base_entry
;
1283 entry
->delta_size
= entry
->size
;
1284 entry
->delta_sibling
= base_entry
->delta_child
;
1285 base_entry
->delta_child
= entry
;
1286 unuse_pack(&w_curs
);
1292 * This must be a delta and we already know what the
1293 * final object type is. Let's extract the actual
1294 * object size from the delta header.
1296 entry
->size
= get_size_from_delta(p
, &w_curs
,
1297 entry
->in_pack_offset
+ entry
->in_pack_header_size
);
1298 if (entry
->size
== 0)
1300 unuse_pack(&w_curs
);
1305 * No choice but to fall back to the recursive delta walk
1306 * with sha1_object_info() to find about the object type
1310 unuse_pack(&w_curs
);
1313 entry
->type
= sha1_object_info(entry
->idx
.sha1
, &entry
->size
);
1315 * The error condition is checked in prepare_pack(). This is
1316 * to permit a missing preferred base object to be ignored
1317 * as a preferred base. Doing so can result in a larger
1318 * pack file, but the transfer will still take place.
1322 static int pack_offset_sort(const void *_a
, const void *_b
)
1324 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1325 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1327 /* avoid filesystem trashing with loose objects */
1328 if (!a
->in_pack
&& !b
->in_pack
)
1329 return hashcmp(a
->idx
.sha1
, b
->idx
.sha1
);
1331 if (a
->in_pack
< b
->in_pack
)
1333 if (a
->in_pack
> b
->in_pack
)
1335 return a
->in_pack_offset
< b
->in_pack_offset
? -1 :
1336 (a
->in_pack_offset
> b
->in_pack_offset
);
1339 static void get_object_details(void)
1342 struct object_entry
**sorted_by_offset
;
1344 sorted_by_offset
= xcalloc(nr_objects
, sizeof(struct object_entry
*));
1345 for (i
= 0; i
< nr_objects
; i
++)
1346 sorted_by_offset
[i
] = objects
+ i
;
1347 qsort(sorted_by_offset
, nr_objects
, sizeof(*sorted_by_offset
), pack_offset_sort
);
1349 for (i
= 0; i
< nr_objects
; i
++) {
1350 struct object_entry
*entry
= sorted_by_offset
[i
];
1351 check_object(entry
);
1352 if (big_file_threshold
< entry
->size
)
1353 entry
->no_try_delta
= 1;
1356 free(sorted_by_offset
);
1360 * We search for deltas in a list sorted by type, by filename hash, and then
1361 * by size, so that we see progressively smaller and smaller files.
1362 * That's because we prefer deltas to be from the bigger file
1363 * to the smaller -- deletes are potentially cheaper, but perhaps
1364 * more importantly, the bigger file is likely the more recent
1365 * one. The deepest deltas are therefore the oldest objects which are
1366 * less susceptible to be accessed often.
1368 static int type_size_sort(const void *_a
, const void *_b
)
1370 const struct object_entry
*a
= *(struct object_entry
**)_a
;
1371 const struct object_entry
*b
= *(struct object_entry
**)_b
;
1373 if (a
->type
> b
->type
)
1375 if (a
->type
< b
->type
)
1377 if (a
->hash
> b
->hash
)
1379 if (a
->hash
< b
->hash
)
1381 if (a
->preferred_base
> b
->preferred_base
)
1383 if (a
->preferred_base
< b
->preferred_base
)
1385 if (a
->size
> b
->size
)
1387 if (a
->size
< b
->size
)
1389 return a
< b
? -1 : (a
> b
); /* newest first */
1393 struct object_entry
*entry
;
1395 struct delta_index
*index
;
1399 static int delta_cacheable(unsigned long src_size
, unsigned long trg_size
,
1400 unsigned long delta_size
)
1402 if (max_delta_cache_size
&& delta_cache_size
+ delta_size
> max_delta_cache_size
)
1405 if (delta_size
< cache_max_small_delta_size
)
1408 /* cache delta, if objects are large enough compared to delta size */
1409 if ((src_size
>> 20) + (trg_size
>> 21) > (delta_size
>> 10))
1417 static pthread_mutex_t read_mutex
;
1418 #define read_lock() pthread_mutex_lock(&read_mutex)
1419 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1421 static pthread_mutex_t cache_mutex
;
1422 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1423 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1425 static pthread_mutex_t progress_mutex
;
1426 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1427 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1431 #define read_lock() (void)0
1432 #define read_unlock() (void)0
1433 #define cache_lock() (void)0
1434 #define cache_unlock() (void)0
1435 #define progress_lock() (void)0
1436 #define progress_unlock() (void)0
1440 static int try_delta(struct unpacked
*trg
, struct unpacked
*src
,
1441 unsigned max_depth
, unsigned long *mem_usage
)
1443 struct object_entry
*trg_entry
= trg
->entry
;
1444 struct object_entry
*src_entry
= src
->entry
;
1445 unsigned long trg_size
, src_size
, delta_size
, sizediff
, max_size
, sz
;
1447 enum object_type type
;
1450 /* Don't bother doing diffs between different types */
1451 if (trg_entry
->type
!= src_entry
->type
)
1455 * We do not bother to try a delta that we discarded on an
1456 * earlier try, but only when reusing delta data. Note that
1457 * src_entry that is marked as the preferred_base should always
1458 * be considered, as even if we produce a suboptimal delta against
1459 * it, we will still save the transfer cost, as we already know
1460 * the other side has it and we won't send src_entry at all.
1462 if (reuse_delta
&& trg_entry
->in_pack
&&
1463 trg_entry
->in_pack
== src_entry
->in_pack
&&
1464 !src_entry
->preferred_base
&&
1465 trg_entry
->in_pack_type
!= OBJ_REF_DELTA
&&
1466 trg_entry
->in_pack_type
!= OBJ_OFS_DELTA
)
1469 /* Let's not bust the allowed depth. */
1470 if (src
->depth
>= max_depth
)
1473 /* Now some size filtering heuristics. */
1474 trg_size
= trg_entry
->size
;
1475 if (!trg_entry
->delta
) {
1476 max_size
= trg_size
/2 - 20;
1479 max_size
= trg_entry
->delta_size
;
1480 ref_depth
= trg
->depth
;
1482 max_size
= (uint64_t)max_size
* (max_depth
- src
->depth
) /
1483 (max_depth
- ref_depth
+ 1);
1486 src_size
= src_entry
->size
;
1487 sizediff
= src_size
< trg_size
? trg_size
- src_size
: 0;
1488 if (sizediff
>= max_size
)
1490 if (trg_size
< src_size
/ 32)
1493 /* Load data if not already done */
1496 trg
->data
= read_sha1_file(trg_entry
->idx
.sha1
, &type
, &sz
);
1499 die("object %s cannot be read",
1500 sha1_to_hex(trg_entry
->idx
.sha1
));
1502 die("object %s inconsistent object length (%lu vs %lu)",
1503 sha1_to_hex(trg_entry
->idx
.sha1
), sz
, trg_size
);
1508 src
->data
= read_sha1_file(src_entry
->idx
.sha1
, &type
, &sz
);
1511 if (src_entry
->preferred_base
) {
1512 static int warned
= 0;
1514 warning("object %s cannot be read",
1515 sha1_to_hex(src_entry
->idx
.sha1
));
1517 * Those objects are not included in the
1518 * resulting pack. Be resilient and ignore
1519 * them if they can't be read, in case the
1520 * pack could be created nevertheless.
1524 die("object %s cannot be read",
1525 sha1_to_hex(src_entry
->idx
.sha1
));
1528 die("object %s inconsistent object length (%lu vs %lu)",
1529 sha1_to_hex(src_entry
->idx
.sha1
), sz
, src_size
);
1533 src
->index
= create_delta_index(src
->data
, src_size
);
1535 static int warned
= 0;
1537 warning("suboptimal pack - out of memory");
1540 *mem_usage
+= sizeof_delta_index(src
->index
);
1543 delta_buf
= create_delta(src
->index
, trg
->data
, trg_size
, &delta_size
, max_size
);
1547 if (trg_entry
->delta
) {
1548 /* Prefer only shallower same-sized deltas. */
1549 if (delta_size
== trg_entry
->delta_size
&&
1550 src
->depth
+ 1 >= trg
->depth
) {
1557 * Handle memory allocation outside of the cache
1558 * accounting lock. Compiler will optimize the strangeness
1559 * away when NO_PTHREADS is defined.
1561 free(trg_entry
->delta_data
);
1563 if (trg_entry
->delta_data
) {
1564 delta_cache_size
-= trg_entry
->delta_size
;
1565 trg_entry
->delta_data
= NULL
;
1567 if (delta_cacheable(src_size
, trg_size
, delta_size
)) {
1568 delta_cache_size
+= delta_size
;
1570 trg_entry
->delta_data
= xrealloc(delta_buf
, delta_size
);
1576 trg_entry
->delta
= src_entry
;
1577 trg_entry
->delta_size
= delta_size
;
1578 trg
->depth
= src
->depth
+ 1;
1583 static unsigned int check_delta_limit(struct object_entry
*me
, unsigned int n
)
1585 struct object_entry
*child
= me
->delta_child
;
1588 unsigned int c
= check_delta_limit(child
, n
+ 1);
1591 child
= child
->delta_sibling
;
1596 static unsigned long free_unpacked(struct unpacked
*n
)
1598 unsigned long freed_mem
= sizeof_delta_index(n
->index
);
1599 free_delta_index(n
->index
);
1602 freed_mem
+= n
->entry
->size
;
1611 static void find_deltas(struct object_entry
**list
, unsigned *list_size
,
1612 int window
, int depth
, unsigned *processed
)
1614 uint32_t i
, idx
= 0, count
= 0;
1615 struct unpacked
*array
;
1616 unsigned long mem_usage
= 0;
1618 array
= xcalloc(window
, sizeof(struct unpacked
));
1621 struct object_entry
*entry
;
1622 struct unpacked
*n
= array
+ idx
;
1623 int j
, max_depth
, best_base
= -1;
1632 if (!entry
->preferred_base
) {
1634 display_progress(progress_state
, *processed
);
1638 mem_usage
-= free_unpacked(n
);
1641 while (window_memory_limit
&&
1642 mem_usage
> window_memory_limit
&&
1644 uint32_t tail
= (idx
+ window
- count
) % window
;
1645 mem_usage
-= free_unpacked(array
+ tail
);
1649 /* We do not compute delta to *create* objects we are not
1652 if (entry
->preferred_base
)
1656 * If the current object is at pack edge, take the depth the
1657 * objects that depend on the current object into account
1658 * otherwise they would become too deep.
1661 if (entry
->delta_child
) {
1662 max_depth
-= check_delta_limit(entry
, 0);
1670 uint32_t other_idx
= idx
+ j
;
1672 if (other_idx
>= window
)
1673 other_idx
-= window
;
1674 m
= array
+ other_idx
;
1677 ret
= try_delta(n
, m
, max_depth
, &mem_usage
);
1681 best_base
= other_idx
;
1685 * If we decided to cache the delta data, then it is best
1686 * to compress it right away. First because we have to do
1687 * it anyway, and doing it here while we're threaded will
1688 * save a lot of time in the non threaded write phase,
1689 * as well as allow for caching more deltas within
1690 * the same cache size limit.
1692 * But only if not writing to stdout, since in that case
1693 * the network is most likely throttling writes anyway,
1694 * and therefore it is best to go to the write phase ASAP
1695 * instead, as we can afford spending more time compressing
1696 * between writes at that moment.
1698 if (entry
->delta_data
&& !pack_to_stdout
) {
1699 entry
->z_delta_size
= do_compress(&entry
->delta_data
,
1702 delta_cache_size
-= entry
->delta_size
;
1703 delta_cache_size
+= entry
->z_delta_size
;
1707 /* if we made n a delta, and if n is already at max
1708 * depth, leaving it in the window is pointless. we
1709 * should evict it first.
1711 if (entry
->delta
&& max_depth
<= n
->depth
)
1715 * Move the best delta base up in the window, after the
1716 * currently deltified object, to keep it longer. It will
1717 * be the first base object to be attempted next.
1720 struct unpacked swap
= array
[best_base
];
1721 int dist
= (window
+ idx
- best_base
) % window
;
1722 int dst
= best_base
;
1724 int src
= (dst
+ 1) % window
;
1725 array
[dst
] = array
[src
];
1733 if (count
+ 1 < window
)
1739 for (i
= 0; i
< window
; ++i
) {
1740 free_delta_index(array
[i
].index
);
1741 free(array
[i
].data
);
1748 static void try_to_free_from_threads(size_t size
)
1751 release_pack_memory(size
, -1);
1755 static try_to_free_t old_try_to_free_routine
;
1758 * The main thread waits on the condition that (at least) one of the workers
1759 * has stopped working (which is indicated in the .working member of
1760 * struct thread_params).
1761 * When a work thread has completed its work, it sets .working to 0 and
1762 * signals the main thread and waits on the condition that .data_ready
1766 struct thread_params
{
1768 struct object_entry
**list
;
1775 pthread_mutex_t mutex
;
1776 pthread_cond_t cond
;
1777 unsigned *processed
;
1780 static pthread_cond_t progress_cond
;
1783 * Mutex and conditional variable can't be statically-initialized on Windows.
1785 static void init_threaded_search(void)
1787 init_recursive_mutex(&read_mutex
);
1788 pthread_mutex_init(&cache_mutex
, NULL
);
1789 pthread_mutex_init(&progress_mutex
, NULL
);
1790 pthread_cond_init(&progress_cond
, NULL
);
1791 old_try_to_free_routine
= set_try_to_free_routine(try_to_free_from_threads
);
1794 static void cleanup_threaded_search(void)
1796 set_try_to_free_routine(old_try_to_free_routine
);
1797 pthread_cond_destroy(&progress_cond
);
1798 pthread_mutex_destroy(&read_mutex
);
1799 pthread_mutex_destroy(&cache_mutex
);
1800 pthread_mutex_destroy(&progress_mutex
);
1803 static void *threaded_find_deltas(void *arg
)
1805 struct thread_params
*me
= arg
;
1807 while (me
->remaining
) {
1808 find_deltas(me
->list
, &me
->remaining
,
1809 me
->window
, me
->depth
, me
->processed
);
1813 pthread_cond_signal(&progress_cond
);
1817 * We must not set ->data_ready before we wait on the
1818 * condition because the main thread may have set it to 1
1819 * before we get here. In order to be sure that new
1820 * work is available if we see 1 in ->data_ready, it
1821 * was initialized to 0 before this thread was spawned
1822 * and we reset it to 0 right away.
1824 pthread_mutex_lock(&me
->mutex
);
1825 while (!me
->data_ready
)
1826 pthread_cond_wait(&me
->cond
, &me
->mutex
);
1828 pthread_mutex_unlock(&me
->mutex
);
1830 /* leave ->working 1 so that this doesn't get more work assigned */
1834 static void ll_find_deltas(struct object_entry
**list
, unsigned list_size
,
1835 int window
, int depth
, unsigned *processed
)
1837 struct thread_params
*p
;
1838 int i
, ret
, active_threads
= 0;
1840 init_threaded_search();
1842 if (!delta_search_threads
) /* --threads=0 means autodetect */
1843 delta_search_threads
= online_cpus();
1844 if (delta_search_threads
<= 1) {
1845 find_deltas(list
, &list_size
, window
, depth
, processed
);
1846 cleanup_threaded_search();
1849 if (progress
> pack_to_stdout
)
1850 fprintf(stderr
, "Delta compression using up to %d threads.\n",
1851 delta_search_threads
);
1852 p
= xcalloc(delta_search_threads
, sizeof(*p
));
1854 /* Partition the work amongst work threads. */
1855 for (i
= 0; i
< delta_search_threads
; i
++) {
1856 unsigned sub_size
= list_size
/ (delta_search_threads
- i
);
1858 /* don't use too small segments or no deltas will be found */
1859 if (sub_size
< 2*window
&& i
+1 < delta_search_threads
)
1862 p
[i
].window
= window
;
1864 p
[i
].processed
= processed
;
1866 p
[i
].data_ready
= 0;
1868 /* try to split chunks on "path" boundaries */
1869 while (sub_size
&& sub_size
< list_size
&&
1870 list
[sub_size
]->hash
&&
1871 list
[sub_size
]->hash
== list
[sub_size
-1]->hash
)
1875 p
[i
].list_size
= sub_size
;
1876 p
[i
].remaining
= sub_size
;
1879 list_size
-= sub_size
;
1882 /* Start work threads. */
1883 for (i
= 0; i
< delta_search_threads
; i
++) {
1884 if (!p
[i
].list_size
)
1886 pthread_mutex_init(&p
[i
].mutex
, NULL
);
1887 pthread_cond_init(&p
[i
].cond
, NULL
);
1888 ret
= pthread_create(&p
[i
].thread
, NULL
,
1889 threaded_find_deltas
, &p
[i
]);
1891 die("unable to create thread: %s", strerror(ret
));
1896 * Now let's wait for work completion. Each time a thread is done
1897 * with its work, we steal half of the remaining work from the
1898 * thread with the largest number of unprocessed objects and give
1899 * it to that newly idle thread. This ensure good load balancing
1900 * until the remaining object list segments are simply too short
1901 * to be worth splitting anymore.
1903 while (active_threads
) {
1904 struct thread_params
*target
= NULL
;
1905 struct thread_params
*victim
= NULL
;
1906 unsigned sub_size
= 0;
1910 for (i
= 0; !target
&& i
< delta_search_threads
; i
++)
1915 pthread_cond_wait(&progress_cond
, &progress_mutex
);
1918 for (i
= 0; i
< delta_search_threads
; i
++)
1919 if (p
[i
].remaining
> 2*window
&&
1920 (!victim
|| victim
->remaining
< p
[i
].remaining
))
1923 sub_size
= victim
->remaining
/ 2;
1924 list
= victim
->list
+ victim
->list_size
- sub_size
;
1925 while (sub_size
&& list
[0]->hash
&&
1926 list
[0]->hash
== list
[-1]->hash
) {
1932 * It is possible for some "paths" to have
1933 * so many objects that no hash boundary
1934 * might be found. Let's just steal the
1935 * exact half in that case.
1937 sub_size
= victim
->remaining
/ 2;
1940 target
->list
= list
;
1941 victim
->list_size
-= sub_size
;
1942 victim
->remaining
-= sub_size
;
1944 target
->list_size
= sub_size
;
1945 target
->remaining
= sub_size
;
1946 target
->working
= 1;
1949 pthread_mutex_lock(&target
->mutex
);
1950 target
->data_ready
= 1;
1951 pthread_cond_signal(&target
->cond
);
1952 pthread_mutex_unlock(&target
->mutex
);
1955 pthread_join(target
->thread
, NULL
);
1956 pthread_cond_destroy(&target
->cond
);
1957 pthread_mutex_destroy(&target
->mutex
);
1961 cleanup_threaded_search();
1966 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1969 static int add_ref_tag(const char *path
, const unsigned char *sha1
, int flag
, void *cb_data
)
1971 unsigned char peeled
[20];
1973 if (!prefixcmp(path
, "refs/tags/") && /* is a tag? */
1974 !peel_ref(path
, peeled
) && /* peelable? */
1975 !is_null_sha1(peeled
) && /* annotated tag? */
1976 locate_object_entry(peeled
)) /* object packed? */
1977 add_object_entry(sha1
, OBJ_TAG
, NULL
, 0);
1981 static void prepare_pack(int window
, int depth
)
1983 struct object_entry
**delta_list
;
1984 uint32_t i
, nr_deltas
;
1987 get_object_details();
1990 * If we're locally repacking then we need to be doubly careful
1991 * from now on in order to make sure no stealth corruption gets
1992 * propagated to the new pack. Clients receiving streamed packs
1993 * should validate everything they get anyway so no need to incur
1994 * the additional cost here in that case.
1996 if (!pack_to_stdout
)
1997 do_check_packed_object_crc
= 1;
1999 if (!nr_objects
|| !window
|| !depth
)
2002 delta_list
= xmalloc(nr_objects
* sizeof(*delta_list
));
2005 for (i
= 0; i
< nr_objects
; i
++) {
2006 struct object_entry
*entry
= objects
+ i
;
2009 /* This happens if we decided to reuse existing
2010 * delta from a pack. "reuse_delta &&" is implied.
2014 if (entry
->size
< 50)
2017 if (entry
->no_try_delta
)
2020 if (!entry
->preferred_base
) {
2022 if (entry
->type
< 0)
2023 die("unable to get type of object %s",
2024 sha1_to_hex(entry
->idx
.sha1
));
2026 if (entry
->type
< 0) {
2028 * This object is not found, but we
2029 * don't have to include it anyway.
2035 delta_list
[n
++] = entry
;
2038 if (nr_deltas
&& n
> 1) {
2039 unsigned nr_done
= 0;
2041 progress_state
= start_progress("Compressing objects",
2043 qsort(delta_list
, n
, sizeof(*delta_list
), type_size_sort
);
2044 ll_find_deltas(delta_list
, n
, window
+1, depth
, &nr_done
);
2045 stop_progress(&progress_state
);
2046 if (nr_done
!= nr_deltas
)
2047 die("inconsistency with delta count");
2052 static int git_pack_config(const char *k
, const char *v
, void *cb
)
2054 if (!strcmp(k
, "pack.window")) {
2055 window
= git_config_int(k
, v
);
2058 if (!strcmp(k
, "pack.windowmemory")) {
2059 window_memory_limit
= git_config_ulong(k
, v
);
2062 if (!strcmp(k
, "pack.depth")) {
2063 depth
= git_config_int(k
, v
);
2066 if (!strcmp(k
, "pack.compression")) {
2067 int level
= git_config_int(k
, v
);
2069 level
= Z_DEFAULT_COMPRESSION
;
2070 else if (level
< 0 || level
> Z_BEST_COMPRESSION
)
2071 die("bad pack compression level %d", level
);
2072 pack_compression_level
= level
;
2073 pack_compression_seen
= 1;
2076 if (!strcmp(k
, "pack.deltacachesize")) {
2077 max_delta_cache_size
= git_config_int(k
, v
);
2080 if (!strcmp(k
, "pack.deltacachelimit")) {
2081 cache_max_small_delta_size
= git_config_int(k
, v
);
2084 if (!strcmp(k
, "pack.threads")) {
2085 delta_search_threads
= git_config_int(k
, v
);
2086 if (delta_search_threads
< 0)
2087 die("invalid number of threads specified (%d)",
2088 delta_search_threads
);
2090 if (delta_search_threads
!= 1)
2091 warning("no threads support, ignoring %s", k
);
2095 if (!strcmp(k
, "pack.indexversion")) {
2096 pack_idx_opts
.version
= git_config_int(k
, v
);
2097 if (pack_idx_opts
.version
> 2)
2098 die("bad pack.indexversion=%"PRIu32
,
2099 pack_idx_opts
.version
);
2102 return git_default_config(k
, v
, cb
);
2105 static void read_object_list_from_stdin(void)
2107 char line
[40 + 1 + PATH_MAX
+ 2];
2108 unsigned char sha1
[20];
2111 if (!fgets(line
, sizeof(line
), stdin
)) {
2115 die("fgets returned NULL, not EOF, not error!");
2121 if (line
[0] == '-') {
2122 if (get_sha1_hex(line
+1, sha1
))
2123 die("expected edge sha1, got garbage:\n %s",
2125 add_preferred_base(sha1
);
2128 if (get_sha1_hex(line
, sha1
))
2129 die("expected sha1, got garbage:\n %s", line
);
2131 add_preferred_base_object(line
+41);
2132 add_object_entry(sha1
, 0, line
+41, 0);
2136 #define OBJECT_ADDED (1u<<20)
2138 static void show_commit(struct commit
*commit
, void *data
)
2140 add_object_entry(commit
->object
.sha1
, OBJ_COMMIT
, NULL
, 0);
2141 commit
->object
.flags
|= OBJECT_ADDED
;
2144 static void show_object(struct object
*obj
,
2145 const struct name_path
*path
, const char *last
,
2148 char *name
= path_name(path
, last
);
2150 add_preferred_base_object(name
);
2151 add_object_entry(obj
->sha1
, obj
->type
, name
, 0);
2152 obj
->flags
|= OBJECT_ADDED
;
2155 * We will have generated the hash from the name,
2156 * but not saved a pointer to it - we can free it
2161 static void show_edge(struct commit
*commit
)
2163 add_preferred_base(commit
->object
.sha1
);
2166 struct in_pack_object
{
2168 struct object
*object
;
2174 struct in_pack_object
*array
;
2177 static void mark_in_pack_object(struct object
*object
, struct packed_git
*p
, struct in_pack
*in_pack
)
2179 in_pack
->array
[in_pack
->nr
].offset
= find_pack_entry_one(object
->sha1
, p
);
2180 in_pack
->array
[in_pack
->nr
].object
= object
;
2185 * Compare the objects in the offset order, in order to emulate the
2186 * "git rev-list --objects" output that produced the pack originally.
2188 static int ofscmp(const void *a_
, const void *b_
)
2190 struct in_pack_object
*a
= (struct in_pack_object
*)a_
;
2191 struct in_pack_object
*b
= (struct in_pack_object
*)b_
;
2193 if (a
->offset
< b
->offset
)
2195 else if (a
->offset
> b
->offset
)
2198 return hashcmp(a
->object
->sha1
, b
->object
->sha1
);
2201 static void add_objects_in_unpacked_packs(struct rev_info
*revs
)
2203 struct packed_git
*p
;
2204 struct in_pack in_pack
;
2207 memset(&in_pack
, 0, sizeof(in_pack
));
2209 for (p
= packed_git
; p
; p
= p
->next
) {
2210 const unsigned char *sha1
;
2213 if (!p
->pack_local
|| p
->pack_keep
)
2215 if (open_pack_index(p
))
2216 die("cannot open pack index");
2218 ALLOC_GROW(in_pack
.array
,
2219 in_pack
.nr
+ p
->num_objects
,
2222 for (i
= 0; i
< p
->num_objects
; i
++) {
2223 sha1
= nth_packed_object_sha1(p
, i
);
2224 o
= lookup_unknown_object(sha1
);
2225 if (!(o
->flags
& OBJECT_ADDED
))
2226 mark_in_pack_object(o
, p
, &in_pack
);
2227 o
->flags
|= OBJECT_ADDED
;
2232 qsort(in_pack
.array
, in_pack
.nr
, sizeof(in_pack
.array
[0]),
2234 for (i
= 0; i
< in_pack
.nr
; i
++) {
2235 struct object
*o
= in_pack
.array
[i
].object
;
2236 add_object_entry(o
->sha1
, o
->type
, "", 0);
2239 free(in_pack
.array
);
2242 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1
)
2244 static struct packed_git
*last_found
= (void *)1;
2245 struct packed_git
*p
;
2247 p
= (last_found
!= (void *)1) ? last_found
: packed_git
;
2250 if ((!p
->pack_local
|| p
->pack_keep
) &&
2251 find_pack_entry_one(sha1
, p
)) {
2255 if (p
== last_found
)
2259 if (p
== last_found
)
2265 static void loosen_unused_packed_objects(struct rev_info
*revs
)
2267 struct packed_git
*p
;
2269 const unsigned char *sha1
;
2271 for (p
= packed_git
; p
; p
= p
->next
) {
2272 if (!p
->pack_local
|| p
->pack_keep
)
2275 if (unpack_unreachable_expiration
&&
2276 p
->mtime
< unpack_unreachable_expiration
)
2279 if (open_pack_index(p
))
2280 die("cannot open pack index");
2282 for (i
= 0; i
< p
->num_objects
; i
++) {
2283 sha1
= nth_packed_object_sha1(p
, i
);
2284 if (!locate_object_entry(sha1
) &&
2285 !has_sha1_pack_kept_or_nonlocal(sha1
))
2286 if (force_object_loose(sha1
, p
->mtime
))
2287 die("unable to force loose object");
2292 static void get_object_list(int ac
, const char **av
)
2294 struct rev_info revs
;
2298 init_revisions(&revs
, NULL
);
2299 save_commit_buffer
= 0;
2300 setup_revisions(ac
, av
, &revs
, NULL
);
2302 while (fgets(line
, sizeof(line
), stdin
) != NULL
) {
2303 int len
= strlen(line
);
2304 if (len
&& line
[len
- 1] == '\n')
2309 if (!strcmp(line
, "--not")) {
2310 flags
^= UNINTERESTING
;
2313 die("not a rev '%s'", line
);
2315 if (handle_revision_arg(line
, &revs
, flags
, 1))
2316 die("bad revision '%s'", line
);
2319 if (prepare_revision_walk(&revs
))
2320 die("revision walk setup failed");
2321 mark_edges_uninteresting(revs
.commits
, &revs
, show_edge
);
2322 traverse_commit_list(&revs
, show_commit
, show_object
, NULL
);
2324 if (keep_unreachable
)
2325 add_objects_in_unpacked_packs(&revs
);
2326 if (unpack_unreachable
)
2327 loosen_unused_packed_objects(&revs
);
2330 static int option_parse_index_version(const struct option
*opt
,
2331 const char *arg
, int unset
)
2334 const char *val
= arg
;
2335 pack_idx_opts
.version
= strtoul(val
, &c
, 10);
2336 if (pack_idx_opts
.version
> 2)
2337 die(_("unsupported index version %s"), val
);
2338 if (*c
== ',' && c
[1])
2339 pack_idx_opts
.off32_limit
= strtoul(c
+1, &c
, 0);
2340 if (*c
|| pack_idx_opts
.off32_limit
& 0x80000000)
2341 die(_("bad index version '%s'"), val
);
2345 static int option_parse_unpack_unreachable(const struct option
*opt
,
2346 const char *arg
, int unset
)
2349 unpack_unreachable
= 0;
2350 unpack_unreachable_expiration
= 0;
2353 unpack_unreachable
= 1;
2355 unpack_unreachable_expiration
= approxidate(arg
);
2360 static int option_parse_ulong(const struct option
*opt
,
2361 const char *arg
, int unset
)
2364 die(_("option %s does not accept negative form"),
2367 if (!git_parse_ulong(arg
, opt
->value
))
2368 die(_("unable to parse value '%s' for option %s"),
2369 arg
, opt
->long_name
);
2373 #define OPT_ULONG(s, l, v, h) \
2374 { OPTION_CALLBACK, (s), (l), (v), "n", (h), \
2375 PARSE_OPT_NONEG, option_parse_ulong }
2377 int cmd_pack_objects(int argc
, const char **argv
, const char *prefix
)
2379 int use_internal_rev_list
= 0;
2381 int all_progress_implied
= 0;
2382 const char *rp_av
[6];
2384 int rev_list_unpacked
= 0, rev_list_all
= 0, rev_list_reflog
= 0;
2385 struct option pack_objects_options
[] = {
2386 OPT_SET_INT('q', "quiet", &progress
,
2387 "do not show progress meter", 0),
2388 OPT_SET_INT(0, "progress", &progress
,
2389 "show progress meter", 1),
2390 OPT_SET_INT(0, "all-progress", &progress
,
2391 "show progress meter during object writing phase", 2),
2392 OPT_BOOL(0, "all-progress-implied",
2393 &all_progress_implied
,
2394 "similar to --all-progress when progress meter is shown"),
2395 { OPTION_CALLBACK
, 0, "index-version", NULL
, "version[,offset]",
2396 "write the pack index file in the specified idx format version",
2397 0, option_parse_index_version
},
2398 OPT_ULONG(0, "max-pack-size", &pack_size_limit
,
2399 "maximum size of each output pack file"),
2400 OPT_BOOL(0, "local", &local
,
2401 "ignore borrowed objects from alternate object store"),
2402 OPT_BOOL(0, "incremental", &incremental
,
2403 "ignore packed objects"),
2404 OPT_INTEGER(0, "window", &window
,
2405 "limit pack window by objects"),
2406 OPT_ULONG(0, "window-memory", &window_memory_limit
,
2407 "limit pack window by memory in addition to object limit"),
2408 OPT_INTEGER(0, "depth", &depth
,
2409 "maximum length of delta chain allowed in the resulting pack"),
2410 OPT_BOOL(0, "reuse-delta", &reuse_delta
,
2411 "reuse existing deltas"),
2412 OPT_BOOL(0, "reuse-object", &reuse_object
,
2413 "reuse existing objects"),
2414 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta
,
2415 "use OFS_DELTA objects"),
2416 OPT_INTEGER(0, "threads", &delta_search_threads
,
2417 "use threads when searching for best delta matches"),
2418 OPT_BOOL(0, "non-empty", &non_empty
,
2419 "do not create an empty pack output"),
2420 OPT_BOOL(0, "revs", &use_internal_rev_list
,
2421 "read revision arguments from standard input"),
2422 { OPTION_SET_INT
, 0, "unpacked", &rev_list_unpacked
, NULL
,
2423 "limit the objects to those that are not yet packed",
2424 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2425 { OPTION_SET_INT
, 0, "all", &rev_list_all
, NULL
,
2426 "include objects reachable from any reference",
2427 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2428 { OPTION_SET_INT
, 0, "reflog", &rev_list_reflog
, NULL
,
2429 "include objects referred by reflog entries",
2430 PARSE_OPT_NOARG
| PARSE_OPT_NONEG
, NULL
, 1 },
2431 OPT_BOOL(0, "stdout", &pack_to_stdout
,
2432 "output pack to stdout"),
2433 OPT_BOOL(0, "include-tag", &include_tag
,
2434 "include tag objects that refer to objects to be packed"),
2435 OPT_BOOL(0, "keep-unreachable", &keep_unreachable
,
2436 "keep unreachable objects"),
2437 { OPTION_CALLBACK
, 0, "unpack-unreachable", NULL
, "time",
2438 "unpack unreachable objects newer than <time>",
2439 PARSE_OPT_OPTARG
, option_parse_unpack_unreachable
},
2440 OPT_BOOL(0, "thin", &thin
,
2441 "create thin packs"),
2442 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep
,
2443 "ignore packs that have companion .keep file"),
2444 OPT_INTEGER(0, "compression", &pack_compression_level
,
2445 "pack compression level"),
2446 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents
,
2447 "do not hide commits by grafts", 0),
2451 read_replace_refs
= 0;
2453 reset_pack_idx_option(&pack_idx_opts
);
2454 git_config(git_pack_config
, NULL
);
2455 if (!pack_compression_seen
&& core_compression_seen
)
2456 pack_compression_level
= core_compression_level
;
2458 progress
= isatty(2);
2459 argc
= parse_options(argc
, argv
, prefix
, pack_objects_options
,
2463 base_name
= argv
[0];
2466 if (pack_to_stdout
!= !base_name
|| argc
)
2467 usage_with_options(pack_usage
, pack_objects_options
);
2469 rp_av
[rp_ac
++] = "pack-objects";
2471 use_internal_rev_list
= 1;
2472 rp_av
[rp_ac
++] = "--objects-edge";
2474 rp_av
[rp_ac
++] = "--objects";
2477 use_internal_rev_list
= 1;
2478 rp_av
[rp_ac
++] = "--all";
2480 if (rev_list_reflog
) {
2481 use_internal_rev_list
= 1;
2482 rp_av
[rp_ac
++] = "--reflog";
2484 if (rev_list_unpacked
) {
2485 use_internal_rev_list
= 1;
2486 rp_av
[rp_ac
++] = "--unpacked";
2491 if (pack_compression_level
== -1)
2492 pack_compression_level
= Z_DEFAULT_COMPRESSION
;
2493 else if (pack_compression_level
< 0 || pack_compression_level
> Z_BEST_COMPRESSION
)
2494 die("bad pack compression level %d", pack_compression_level
);
2496 if (delta_search_threads
!= 1)
2497 warning("no threads support, ignoring --threads");
2499 if (!pack_to_stdout
&& !pack_size_limit
)
2500 pack_size_limit
= pack_size_limit_cfg
;
2501 if (pack_to_stdout
&& pack_size_limit
)
2502 die("--max-pack-size cannot be used to build a pack for transfer.");
2503 if (pack_size_limit
&& pack_size_limit
< 1024*1024) {
2504 warning("minimum pack size limit is 1 MiB");
2505 pack_size_limit
= 1024*1024;
2508 if (!pack_to_stdout
&& thin
)
2509 die("--thin cannot be used to build an indexable pack.");
2511 if (keep_unreachable
&& unpack_unreachable
)
2512 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2514 if (progress
&& all_progress_implied
)
2517 prepare_packed_git();
2520 progress_state
= start_progress("Counting objects", 0);
2521 if (!use_internal_rev_list
)
2522 read_object_list_from_stdin();
2524 rp_av
[rp_ac
] = NULL
;
2525 get_object_list(rp_ac
, rp_av
);
2527 cleanup_preferred_base();
2528 if (include_tag
&& nr_result
)
2529 for_each_ref(add_ref_tag
, NULL
);
2530 stop_progress(&progress_state
);
2532 if (non_empty
&& !nr_result
)
2535 prepare_pack(window
, depth
);
2538 fprintf(stderr
, "Total %"PRIu32
" (delta %"PRIu32
"),"
2539 " reused %"PRIu32
" (delta %"PRIu32
")\n",
2540 written
, written_delta
, reused
, reused_delta
);