Merge branch 'sg/index-format-doc-update' into maint
[git/debian.git] / hash-lookup.c
blobb98ed5e11e8a3af02a99b6d2a977cc804a122414
1 #include "cache.h"
2 #include "hash-lookup.h"
4 static uint32_t take2(const struct object_id *oid, size_t ofs)
6 return ((oid->hash[ofs] << 8) | oid->hash[ofs + 1]);
9 /*
10 * Conventional binary search loop looks like this:
12 * do {
13 * int mi = lo + (hi - lo) / 2;
14 * int cmp = "entry pointed at by mi" minus "target";
15 * if (!cmp)
16 * return (mi is the wanted one)
17 * if (cmp > 0)
18 * hi = mi; "mi is larger than target"
19 * else
20 * lo = mi+1; "mi is smaller than target"
21 * } while (lo < hi);
23 * The invariants are:
25 * - When entering the loop, lo points at a slot that is never
26 * above the target (it could be at the target), hi points at a
27 * slot that is guaranteed to be above the target (it can never
28 * be at the target).
30 * - We find a point 'mi' between lo and hi (mi could be the same
31 * as lo, but never can be the same as hi), and check if it hits
32 * the target. There are three cases:
34 * - if it is a hit, we are happy.
36 * - if it is strictly higher than the target, we update hi with
37 * it.
39 * - if it is strictly lower than the target, we update lo to be
40 * one slot after it, because we allow lo to be at the target.
42 * When choosing 'mi', we do not have to take the "middle" but
43 * anywhere in between lo and hi, as long as lo <= mi < hi is
44 * satisfied. When we somehow know that the distance between the
45 * target and lo is much shorter than the target and hi, we could
46 * pick mi that is much closer to lo than the midway.
49 * The table should contain "nr" elements.
50 * The oid of element i (between 0 and nr - 1) should be returned
51 * by "fn(i, table)".
53 int oid_pos(const struct object_id *oid, const void *table, size_t nr,
54 oid_access_fn fn)
56 size_t hi = nr;
57 size_t lo = 0;
58 size_t mi = 0;
60 if (!nr)
61 return -1;
63 if (nr != 1) {
64 size_t lov, hiv, miv, ofs;
66 for (ofs = 0; ofs < the_hash_algo->rawsz - 2; ofs += 2) {
67 lov = take2(fn(0, table), ofs);
68 hiv = take2(fn(nr - 1, table), ofs);
69 miv = take2(oid, ofs);
70 if (miv < lov)
71 return -1;
72 if (hiv < miv)
73 return index_pos_to_insert_pos(nr);
74 if (lov != hiv) {
76 * At this point miv could be equal
77 * to hiv (but hash could still be higher);
78 * the invariant of (mi < hi) should be
79 * kept.
81 mi = (nr - 1) * (miv - lov) / (hiv - lov);
82 if (lo <= mi && mi < hi)
83 break;
84 BUG("assertion failed in binary search");
89 do {
90 int cmp;
91 cmp = oidcmp(fn(mi, table), oid);
92 if (!cmp)
93 return mi;
94 if (cmp > 0)
95 hi = mi;
96 else
97 lo = mi + 1;
98 mi = lo + (hi - lo) / 2;
99 } while (lo < hi);
100 return index_pos_to_insert_pos(lo);
103 int bsearch_hash(const unsigned char *hash, const uint32_t *fanout_nbo,
104 const unsigned char *table, size_t stride, uint32_t *result)
106 uint32_t hi, lo;
108 hi = ntohl(fanout_nbo[*hash]);
109 lo = ((*hash == 0x0) ? 0 : ntohl(fanout_nbo[*hash - 1]));
111 while (lo < hi) {
112 unsigned mi = lo + (hi - lo) / 2;
113 int cmp = hashcmp(table + mi * stride, hash);
115 if (!cmp) {
116 if (result)
117 *result = mi;
118 return 1;
120 if (cmp > 0)
121 hi = mi;
122 else
123 lo = mi + 1;
126 if (result)
127 *result = lo;
128 return 0;