2 #include "pack-revindex.h"
3 #include "object-store.h"
6 * Pack index for existing packs give us easy access to the offsets into
7 * corresponding pack file where each object's data starts, but the entries
8 * do not store the size of the compressed representation (uncompressed
9 * size is easily available by examining the pack entry header). It is
10 * also rather expensive to find the sha1 for an object given its offset.
12 * The pack index file is sorted by object name mapping to offset;
13 * this revindex array is a list of offset/index_nr pairs
14 * ordered by offset, so if you know the offset of an object, next offset
15 * is where its packed representation ends and the index_nr can be used to
16 * get the object sha1 from the main index.
20 * This is a least-significant-digit radix sort.
22 * It sorts each of the "n" items in "entries" by its offset field. The "max"
23 * parameter must be at least as large as the largest offset in the array,
24 * and lets us quit the sort early.
26 static void sort_revindex(struct revindex_entry
*entries
, unsigned n
, off_t max
)
29 * We use a "digit" size of 16 bits. That keeps our memory
30 * usage reasonable, and we can generally (for a 4G or smaller
31 * packfile) quit after two rounds of radix-sorting.
33 #define DIGIT_SIZE (16)
34 #define BUCKETS (1 << DIGIT_SIZE)
36 * We want to know the bucket that a[i] will go into when we are using
37 * the digit that is N bits from the (least significant) end.
39 #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
42 * We need O(n) temporary storage. Rather than do an extra copy of the
43 * partial results into "entries", we sort back and forth between the
44 * real array and temporary storage. In each iteration of the loop, we
45 * keep track of them with alias pointers, always sorting from "from"
48 struct revindex_entry
*tmp
, *from
, *to
;
52 ALLOC_ARRAY(pos
, BUCKETS
);
58 * If (max >> bits) is zero, then we know that the radix digit we are
59 * on (and any higher) will be zero for all entries, and our loop will
60 * be a no-op, as everybody lands in the same zero-th bucket.
62 for (bits
= 0; max
>> bits
; bits
+= DIGIT_SIZE
) {
65 memset(pos
, 0, BUCKETS
* sizeof(*pos
));
68 * We want pos[i] to store the index of the last element that
69 * will go in bucket "i" (actually one past the last element).
70 * To do this, we first count the items that will go in each
71 * bucket, which gives us a relative offset from the last
72 * bucket. We can then cumulatively add the index from the
73 * previous bucket to get the true index.
75 for (i
= 0; i
< n
; i
++)
76 pos
[BUCKET_FOR(from
, i
, bits
)]++;
77 for (i
= 1; i
< BUCKETS
; i
++)
81 * Now we can drop the elements into their correct buckets (in
82 * our temporary array). We iterate the pos counter backwards
83 * to avoid using an extra index to count up. And since we are
84 * going backwards there, we must also go backwards through the
85 * array itself, to keep the sort stable.
87 * Note that we use an unsigned iterator to make sure we can
88 * handle 2^32-1 objects, even on a 32-bit system. But this
89 * means we cannot use the more obvious "i >= 0" loop condition
90 * for counting backwards, and must instead check for
91 * wrap-around with UINT_MAX.
93 for (i
= n
- 1; i
!= UINT_MAX
; i
--)
94 to
[--pos
[BUCKET_FOR(from
, i
, bits
)]] = from
[i
];
97 * Now "to" contains the most sorted list, so we swap "from" and
98 * "to" for the next iteration.
104 * If we ended with our data in the original array, great. If not,
105 * we have to move it back from the temporary storage.
108 COPY_ARRAY(entries
, tmp
, n
);
118 * Ordered list of offsets of objects in the pack.
120 static void create_pack_revindex(struct packed_git
*p
)
122 unsigned num_ent
= p
->num_objects
;
124 const char *index
= p
->index_data
;
125 const unsigned hashsz
= the_hash_algo
->rawsz
;
127 ALLOC_ARRAY(p
->revindex
, num_ent
+ 1);
130 if (p
->index_version
> 1) {
131 const uint32_t *off_32
=
132 (uint32_t *)(index
+ 8 + p
->num_objects
* (hashsz
+ 4));
133 const uint32_t *off_64
= off_32
+ p
->num_objects
;
134 for (i
= 0; i
< num_ent
; i
++) {
135 uint32_t off
= ntohl(*off_32
++);
136 if (!(off
& 0x80000000)) {
137 p
->revindex
[i
].offset
= off
;
139 p
->revindex
[i
].offset
= get_be64(off_64
);
142 p
->revindex
[i
].nr
= i
;
145 for (i
= 0; i
< num_ent
; i
++) {
146 uint32_t hl
= *((uint32_t *)(index
+ (hashsz
+ 4) * i
));
147 p
->revindex
[i
].offset
= ntohl(hl
);
148 p
->revindex
[i
].nr
= i
;
153 * This knows the pack format -- the hash trailer
154 * follows immediately after the last object data.
156 p
->revindex
[num_ent
].offset
= p
->pack_size
- hashsz
;
157 p
->revindex
[num_ent
].nr
= -1;
158 sort_revindex(p
->revindex
, num_ent
, p
->pack_size
);
161 void load_pack_revindex(struct packed_git
*p
)
164 create_pack_revindex(p
);
167 int find_revindex_position(struct packed_git
*p
, off_t ofs
)
170 int hi
= p
->num_objects
+ 1;
171 struct revindex_entry
*revindex
= p
->revindex
;
174 unsigned mi
= lo
+ (hi
- lo
) / 2;
175 if (revindex
[mi
].offset
== ofs
) {
177 } else if (ofs
< revindex
[mi
].offset
)
183 error("bad offset for revindex");
187 struct revindex_entry
*find_pack_revindex(struct packed_git
*p
, off_t ofs
)
191 load_pack_revindex(p
);
192 pos
= find_revindex_position(p
, ofs
);
197 return p
->revindex
+ pos
;