The first batch post 2.24 cycle
[git.git] / pack-revindex.c
blobd28a7e43d0bd80c3dc968439b3262da7c80117c1
1 #include "cache.h"
2 #include "pack-revindex.h"
3 #include "object-store.h"
4 #include "packfile.h"
6 /*
7 * Pack index for existing packs give us easy access to the offsets into
8 * corresponding pack file where each object's data starts, but the entries
9 * do not store the size of the compressed representation (uncompressed
10 * size is easily available by examining the pack entry header). It is
11 * also rather expensive to find the sha1 for an object given its offset.
13 * The pack index file is sorted by object name mapping to offset;
14 * this revindex array is a list of offset/index_nr pairs
15 * ordered by offset, so if you know the offset of an object, next offset
16 * is where its packed representation ends and the index_nr can be used to
17 * get the object sha1 from the main index.
21 * This is a least-significant-digit radix sort.
23 * It sorts each of the "n" items in "entries" by its offset field. The "max"
24 * parameter must be at least as large as the largest offset in the array,
25 * and lets us quit the sort early.
27 static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
30 * We use a "digit" size of 16 bits. That keeps our memory
31 * usage reasonable, and we can generally (for a 4G or smaller
32 * packfile) quit after two rounds of radix-sorting.
34 #define DIGIT_SIZE (16)
35 #define BUCKETS (1 << DIGIT_SIZE)
37 * We want to know the bucket that a[i] will go into when we are using
38 * the digit that is N bits from the (least significant) end.
40 #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
43 * We need O(n) temporary storage. Rather than do an extra copy of the
44 * partial results into "entries", we sort back and forth between the
45 * real array and temporary storage. In each iteration of the loop, we
46 * keep track of them with alias pointers, always sorting from "from"
47 * to "to".
49 struct revindex_entry *tmp, *from, *to;
50 int bits;
51 unsigned *pos;
53 ALLOC_ARRAY(pos, BUCKETS);
54 ALLOC_ARRAY(tmp, n);
55 from = entries;
56 to = tmp;
59 * If (max >> bits) is zero, then we know that the radix digit we are
60 * on (and any higher) will be zero for all entries, and our loop will
61 * be a no-op, as everybody lands in the same zero-th bucket.
63 for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
64 unsigned i;
66 memset(pos, 0, BUCKETS * sizeof(*pos));
69 * We want pos[i] to store the index of the last element that
70 * will go in bucket "i" (actually one past the last element).
71 * To do this, we first count the items that will go in each
72 * bucket, which gives us a relative offset from the last
73 * bucket. We can then cumulatively add the index from the
74 * previous bucket to get the true index.
76 for (i = 0; i < n; i++)
77 pos[BUCKET_FOR(from, i, bits)]++;
78 for (i = 1; i < BUCKETS; i++)
79 pos[i] += pos[i-1];
82 * Now we can drop the elements into their correct buckets (in
83 * our temporary array). We iterate the pos counter backwards
84 * to avoid using an extra index to count up. And since we are
85 * going backwards there, we must also go backwards through the
86 * array itself, to keep the sort stable.
88 * Note that we use an unsigned iterator to make sure we can
89 * handle 2^32-1 objects, even on a 32-bit system. But this
90 * means we cannot use the more obvious "i >= 0" loop condition
91 * for counting backwards, and must instead check for
92 * wrap-around with UINT_MAX.
94 for (i = n - 1; i != UINT_MAX; i--)
95 to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
98 * Now "to" contains the most sorted list, so we swap "from" and
99 * "to" for the next iteration.
101 SWAP(from, to);
105 * If we ended with our data in the original array, great. If not,
106 * we have to move it back from the temporary storage.
108 if (from != entries)
109 COPY_ARRAY(entries, tmp, n);
110 free(tmp);
111 free(pos);
113 #undef BUCKET_FOR
114 #undef BUCKETS
115 #undef DIGIT_SIZE
119 * Ordered list of offsets of objects in the pack.
121 static void create_pack_revindex(struct packed_git *p)
123 const unsigned num_ent = p->num_objects;
124 unsigned i;
125 const char *index = p->index_data;
126 const unsigned hashsz = the_hash_algo->rawsz;
128 ALLOC_ARRAY(p->revindex, num_ent + 1);
129 index += 4 * 256;
131 if (p->index_version > 1) {
132 const uint32_t *off_32 =
133 (uint32_t *)(index + 8 + p->num_objects * (hashsz + 4));
134 const uint32_t *off_64 = off_32 + p->num_objects;
135 for (i = 0; i < num_ent; i++) {
136 const uint32_t off = ntohl(*off_32++);
137 if (!(off & 0x80000000)) {
138 p->revindex[i].offset = off;
139 } else {
140 p->revindex[i].offset = get_be64(off_64);
141 off_64 += 2;
143 p->revindex[i].nr = i;
145 } else {
146 for (i = 0; i < num_ent; i++) {
147 const uint32_t hl = *((uint32_t *)(index + (hashsz + 4) * i));
148 p->revindex[i].offset = ntohl(hl);
149 p->revindex[i].nr = i;
154 * This knows the pack format -- the hash trailer
155 * follows immediately after the last object data.
157 p->revindex[num_ent].offset = p->pack_size - hashsz;
158 p->revindex[num_ent].nr = -1;
159 sort_revindex(p->revindex, num_ent, p->pack_size);
162 int load_pack_revindex(struct packed_git *p)
164 if (!p->revindex) {
165 if (open_pack_index(p))
166 return -1;
167 create_pack_revindex(p);
169 return 0;
172 int find_revindex_position(struct packed_git *p, off_t ofs)
174 int lo = 0;
175 int hi = p->num_objects + 1;
176 const struct revindex_entry *revindex = p->revindex;
178 do {
179 const unsigned mi = lo + (hi - lo) / 2;
180 if (revindex[mi].offset == ofs) {
181 return mi;
182 } else if (ofs < revindex[mi].offset)
183 hi = mi;
184 else
185 lo = mi + 1;
186 } while (lo < hi);
188 error("bad offset for revindex");
189 return -1;
192 struct revindex_entry *find_pack_revindex(struct packed_git *p, off_t ofs)
194 int pos;
196 if (load_pack_revindex(p))
197 return NULL;
199 pos = find_revindex_position(p, ofs);
201 if (pos < 0)
202 return NULL;
204 return p->revindex + pos;