use SWAP macro
[git/git.git] / pack-revindex.c
CommitLineData
3449f8c4
NP
1#include "cache.h"
2#include "pack-revindex.h"
3
4/*
5 * Pack index for existing packs give us easy access to the offsets into
6 * corresponding pack file where each object's data starts, but the entries
7 * do not store the size of the compressed representation (uncompressed
8 * size is easily available by examining the pack entry header). It is
9 * also rather expensive to find the sha1 for an object given its offset.
10 *
f4015337
JK
11 * The pack index file is sorted by object name mapping to offset;
12 * this revindex array is a list of offset/index_nr pairs
3449f8c4
NP
13 * ordered by offset, so if you know the offset of an object, next offset
14 * is where its packed representation ends and the index_nr can be used to
15 * get the object sha1 from the main index.
16 */
17
8b8dfd51
JK
18/*
19 * This is a least-significant-digit radix sort.
20 *
21 * It sorts each of the "n" items in "entries" by its offset field. The "max"
22 * parameter must be at least as large as the largest offset in the array,
23 * and lets us quit the sort early.
24 */
25static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
3449f8c4 26{
8b8dfd51
JK
27 /*
28 * We use a "digit" size of 16 bits. That keeps our memory
29 * usage reasonable, and we can generally (for a 4G or smaller
30 * packfile) quit after two rounds of radix-sorting.
31 */
32#define DIGIT_SIZE (16)
33#define BUCKETS (1 << DIGIT_SIZE)
34 /*
35 * We want to know the bucket that a[i] will go into when we are using
36 * the digit that is N bits from the (least significant) end.
37 */
38#define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
39
40 /*
41 * We need O(n) temporary storage. Rather than do an extra copy of the
42 * partial results into "entries", we sort back and forth between the
43 * real array and temporary storage. In each iteration of the loop, we
44 * keep track of them with alias pointers, always sorting from "from"
45 * to "to".
46 */
b32fa95f 47 struct revindex_entry *tmp, *from, *to;
8b8dfd51 48 int bits;
b32fa95f
JK
49 unsigned *pos;
50
51 ALLOC_ARRAY(pos, BUCKETS);
52 ALLOC_ARRAY(tmp, n);
53 from = entries;
54 to = tmp;
8b8dfd51
JK
55
56 /*
57 * If (max >> bits) is zero, then we know that the radix digit we are
58 * on (and any higher) will be zero for all entries, and our loop will
59 * be a no-op, as everybody lands in the same zero-th bucket.
60 */
61 for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
8b8dfd51
JK
62 unsigned i;
63
64 memset(pos, 0, BUCKETS * sizeof(*pos));
65
66 /*
67 * We want pos[i] to store the index of the last element that
68 * will go in bucket "i" (actually one past the last element).
69 * To do this, we first count the items that will go in each
70 * bucket, which gives us a relative offset from the last
71 * bucket. We can then cumulatively add the index from the
72 * previous bucket to get the true index.
73 */
74 for (i = 0; i < n; i++)
75 pos[BUCKET_FOR(from, i, bits)]++;
76 for (i = 1; i < BUCKETS; i++)
77 pos[i] += pos[i-1];
78
79 /*
80 * Now we can drop the elements into their correct buckets (in
81 * our temporary array). We iterate the pos counter backwards
82 * to avoid using an extra index to count up. And since we are
83 * going backwards there, we must also go backwards through the
84 * array itself, to keep the sort stable.
85 *
86 * Note that we use an unsigned iterator to make sure we can
87 * handle 2^32-1 objects, even on a 32-bit system. But this
88 * means we cannot use the more obvious "i >= 0" loop condition
89 * for counting backwards, and must instead check for
90 * wrap-around with UINT_MAX.
91 */
92 for (i = n - 1; i != UINT_MAX; i--)
93 to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
94
95 /*
96 * Now "to" contains the most sorted list, so we swap "from" and
97 * "to" for the next iteration.
98 */
35d803bc 99 SWAP(from, to);
8b8dfd51
JK
100 }
101
102 /*
103 * If we ended with our data in the original array, great. If not,
104 * we have to move it back from the temporary storage.
105 */
106 if (from != entries)
45ccef87 107 COPY_ARRAY(entries, tmp, n);
8b8dfd51
JK
108 free(tmp);
109 free(pos);
110
111#undef BUCKET_FOR
112#undef BUCKETS
113#undef DIGIT_SIZE
3449f8c4
NP
114}
115
116/*
117 * Ordered list of offsets of objects in the pack.
118 */
9d98bbf5 119static void create_pack_revindex(struct packed_git *p)
3449f8c4 120{
012b32bb
JK
121 unsigned num_ent = p->num_objects;
122 unsigned i;
3449f8c4
NP
123 const char *index = p->index_data;
124
11529ece 125 ALLOC_ARRAY(p->revindex, num_ent + 1);
3449f8c4
NP
126 index += 4 * 256;
127
128 if (p->index_version > 1) {
129 const uint32_t *off_32 =
130 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
131 const uint32_t *off_64 = off_32 + p->num_objects;
132 for (i = 0; i < num_ent; i++) {
133 uint32_t off = ntohl(*off_32++);
134 if (!(off & 0x80000000)) {
9d98bbf5 135 p->revindex[i].offset = off;
3449f8c4 136 } else {
9d98bbf5 137 p->revindex[i].offset =
3449f8c4 138 ((uint64_t)ntohl(*off_64++)) << 32;
9d98bbf5 139 p->revindex[i].offset |=
3449f8c4
NP
140 ntohl(*off_64++);
141 }
9d98bbf5 142 p->revindex[i].nr = i;
3449f8c4
NP
143 }
144 } else {
145 for (i = 0; i < num_ent; i++) {
146 uint32_t hl = *((uint32_t *)(index + 24 * i));
9d98bbf5
JK
147 p->revindex[i].offset = ntohl(hl);
148 p->revindex[i].nr = i;
3449f8c4
NP
149 }
150 }
151
152 /* This knows the pack format -- the 20-byte trailer
153 * follows immediately after the last object data.
154 */
9d98bbf5
JK
155 p->revindex[num_ent].offset = p->pack_size - 20;
156 p->revindex[num_ent].nr = -1;
157 sort_revindex(p->revindex, num_ent, p->pack_size);
3449f8c4
NP
158}
159
9d98bbf5 160void load_pack_revindex(struct packed_git *p)
3449f8c4 161{
9d98bbf5
JK
162 if (!p->revindex)
163 create_pack_revindex(p);
92e5c77c
VM
164}
165
9d98bbf5 166int find_revindex_position(struct packed_git *p, off_t ofs)
92e5c77c
VM
167{
168 int lo = 0;
9d98bbf5
JK
169 int hi = p->num_objects + 1;
170 struct revindex_entry *revindex = p->revindex;
92e5c77c 171
3449f8c4 172 do {
012b32bb 173 unsigned mi = lo + (hi - lo) / 2;
3449f8c4 174 if (revindex[mi].offset == ofs) {
92e5c77c 175 return mi;
3449f8c4
NP
176 } else if (ofs < revindex[mi].offset)
177 hi = mi;
178 else
179 lo = mi + 1;
180 } while (lo < hi);
92e5c77c 181
08698b1e 182 error("bad offset for revindex");
92e5c77c
VM
183 return -1;
184}
185
186struct revindex_entry *find_pack_revindex(struct packed_git *p, off_t ofs)
187{
9d98bbf5
JK
188 int pos;
189
190 load_pack_revindex(p);
191 pos = find_revindex_position(p, ofs);
92e5c77c
VM
192
193 if (pos < 0)
194 return NULL;
195
9d98bbf5 196 return p->revindex + pos;
3449f8c4 197}