Sync with 2.43.4
[git.git] / hash-lookup.c
blob9f0f95e2b9e0c44e79bfe6ee3ec43b61e7991f4a
1 #include "git-compat-util.h"
2 #include "hash.h"
3 #include "hash-lookup.h"
4 #include "read-cache-ll.h"
6 static uint32_t take2(const struct object_id *oid, size_t ofs)
8 return ((oid->hash[ofs] << 8) | oid->hash[ofs + 1]);
12 * Conventional binary search loop looks like this:
14 * do {
15 * int mi = lo + (hi - lo) / 2;
16 * int cmp = "entry pointed at by mi" minus "target";
17 * if (!cmp)
18 * return (mi is the wanted one)
19 * if (cmp > 0)
20 * hi = mi; "mi is larger than target"
21 * else
22 * lo = mi+1; "mi is smaller than target"
23 * } while (lo < hi);
25 * The invariants are:
27 * - When entering the loop, lo points at a slot that is never
28 * above the target (it could be at the target), hi points at a
29 * slot that is guaranteed to be above the target (it can never
30 * be at the target).
32 * - We find a point 'mi' between lo and hi (mi could be the same
33 * as lo, but never can be the same as hi), and check if it hits
34 * the target. There are three cases:
36 * - if it is a hit, we are happy.
38 * - if it is strictly higher than the target, we update hi with
39 * it.
41 * - if it is strictly lower than the target, we update lo to be
42 * one slot after it, because we allow lo to be at the target.
44 * When choosing 'mi', we do not have to take the "middle" but
45 * anywhere in between lo and hi, as long as lo <= mi < hi is
46 * satisfied. When we somehow know that the distance between the
47 * target and lo is much shorter than the target and hi, we could
48 * pick mi that is much closer to lo than the midway.
51 * The table should contain "nr" elements.
52 * The oid of element i (between 0 and nr - 1) should be returned
53 * by "fn(i, table)".
55 int oid_pos(const struct object_id *oid, const void *table, size_t nr,
56 oid_access_fn fn)
58 size_t hi = nr;
59 size_t lo = 0;
60 size_t mi = 0;
62 if (!nr)
63 return -1;
65 if (nr != 1) {
66 size_t lov, hiv, miv, ofs;
68 for (ofs = 0; ofs < the_hash_algo->rawsz - 2; ofs += 2) {
69 lov = take2(fn(0, table), ofs);
70 hiv = take2(fn(nr - 1, table), ofs);
71 miv = take2(oid, ofs);
72 if (miv < lov)
73 return -1;
74 if (hiv < miv)
75 return index_pos_to_insert_pos(nr);
76 if (lov != hiv) {
78 * At this point miv could be equal
79 * to hiv (but hash could still be higher);
80 * the invariant of (mi < hi) should be
81 * kept.
83 mi = (nr - 1) * (miv - lov) / (hiv - lov);
84 if (lo <= mi && mi < hi)
85 break;
86 BUG("assertion failed in binary search");
91 do {
92 int cmp;
93 cmp = oidcmp(fn(mi, table), oid);
94 if (!cmp)
95 return mi;
96 if (cmp > 0)
97 hi = mi;
98 else
99 lo = mi + 1;
100 mi = lo + (hi - lo) / 2;
101 } while (lo < hi);
102 return index_pos_to_insert_pos(lo);
105 int bsearch_hash(const unsigned char *hash, const uint32_t *fanout_nbo,
106 const unsigned char *table, size_t stride, uint32_t *result)
108 uint32_t hi, lo;
110 hi = ntohl(fanout_nbo[*hash]);
111 lo = ((*hash == 0x0) ? 0 : ntohl(fanout_nbo[*hash - 1]));
113 while (lo < hi) {
114 unsigned mi = lo + (hi - lo) / 2;
115 int cmp = hashcmp(table + mi * stride, hash);
117 if (!cmp) {
118 if (result)
119 *result = mi;
120 return 1;
122 if (cmp > 0)
123 hi = mi;
124 else
125 lo = mi + 1;
128 if (result)
129 *result = lo;
130 return 0;