pack-revindex: store entries directly in packed_git
[git/git.git] / pack-revindex.c
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 *
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
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
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 */
25 static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
26 {
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 */
47 struct revindex_entry *tmp = xmalloc(n * sizeof(*tmp));
48 struct revindex_entry *from = entries, *to = tmp;
49 int bits;
50 unsigned *pos = xmalloc(BUCKETS * sizeof(*pos));
51
52 /*
53 * If (max >> bits) is zero, then we know that the radix digit we are
54 * on (and any higher) will be zero for all entries, and our loop will
55 * be a no-op, as everybody lands in the same zero-th bucket.
56 */
57 for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
58 struct revindex_entry *swap;
59 unsigned i;
60
61 memset(pos, 0, BUCKETS * sizeof(*pos));
62
63 /*
64 * We want pos[i] to store the index of the last element that
65 * will go in bucket "i" (actually one past the last element).
66 * To do this, we first count the items that will go in each
67 * bucket, which gives us a relative offset from the last
68 * bucket. We can then cumulatively add the index from the
69 * previous bucket to get the true index.
70 */
71 for (i = 0; i < n; i++)
72 pos[BUCKET_FOR(from, i, bits)]++;
73 for (i = 1; i < BUCKETS; i++)
74 pos[i] += pos[i-1];
75
76 /*
77 * Now we can drop the elements into their correct buckets (in
78 * our temporary array). We iterate the pos counter backwards
79 * to avoid using an extra index to count up. And since we are
80 * going backwards there, we must also go backwards through the
81 * array itself, to keep the sort stable.
82 *
83 * Note that we use an unsigned iterator to make sure we can
84 * handle 2^32-1 objects, even on a 32-bit system. But this
85 * means we cannot use the more obvious "i >= 0" loop condition
86 * for counting backwards, and must instead check for
87 * wrap-around with UINT_MAX.
88 */
89 for (i = n - 1; i != UINT_MAX; i--)
90 to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
91
92 /*
93 * Now "to" contains the most sorted list, so we swap "from" and
94 * "to" for the next iteration.
95 */
96 swap = from;
97 from = to;
98 to = swap;
99 }
100
101 /*
102 * If we ended with our data in the original array, great. If not,
103 * we have to move it back from the temporary storage.
104 */
105 if (from != entries)
106 memcpy(entries, tmp, n * sizeof(*entries));
107 free(tmp);
108 free(pos);
109
110 #undef BUCKET_FOR
111 #undef BUCKETS
112 #undef DIGIT_SIZE
113 }
114
115 /*
116 * Ordered list of offsets of objects in the pack.
117 */
118 static void create_pack_revindex(struct packed_git *p)
119 {
120 unsigned num_ent = p->num_objects;
121 unsigned i;
122 const char *index = p->index_data;
123
124 p->revindex = xmalloc(sizeof(*p->revindex) * (num_ent + 1));
125 index += 4 * 256;
126
127 if (p->index_version > 1) {
128 const uint32_t *off_32 =
129 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
130 const uint32_t *off_64 = off_32 + p->num_objects;
131 for (i = 0; i < num_ent; i++) {
132 uint32_t off = ntohl(*off_32++);
133 if (!(off & 0x80000000)) {
134 p->revindex[i].offset = off;
135 } else {
136 p->revindex[i].offset =
137 ((uint64_t)ntohl(*off_64++)) << 32;
138 p->revindex[i].offset |=
139 ntohl(*off_64++);
140 }
141 p->revindex[i].nr = i;
142 }
143 } else {
144 for (i = 0; i < num_ent; i++) {
145 uint32_t hl = *((uint32_t *)(index + 24 * i));
146 p->revindex[i].offset = ntohl(hl);
147 p->revindex[i].nr = i;
148 }
149 }
150
151 /* This knows the pack format -- the 20-byte trailer
152 * follows immediately after the last object data.
153 */
154 p->revindex[num_ent].offset = p->pack_size - 20;
155 p->revindex[num_ent].nr = -1;
156 sort_revindex(p->revindex, num_ent, p->pack_size);
157 }
158
159 void load_pack_revindex(struct packed_git *p)
160 {
161 if (!p->revindex)
162 create_pack_revindex(p);
163 }
164
165 int find_revindex_position(struct packed_git *p, off_t ofs)
166 {
167 int lo = 0;
168 int hi = p->num_objects + 1;
169 struct revindex_entry *revindex = p->revindex;
170
171 do {
172 unsigned mi = lo + (hi - lo) / 2;
173 if (revindex[mi].offset == ofs) {
174 return mi;
175 } else if (ofs < revindex[mi].offset)
176 hi = mi;
177 else
178 lo = mi + 1;
179 } while (lo < hi);
180
181 error("bad offset for revindex");
182 return -1;
183 }
184
185 struct revindex_entry *find_pack_revindex(struct packed_git *p, off_t ofs)
186 {
187 int pos;
188
189 load_pack_revindex(p);
190 pos = find_revindex_position(p, ofs);
191
192 if (pos < 0)
193 return NULL;
194
195 return p->revindex + pos;
196 }