sha1-lookup: add new "sha1_pos" function to efficiently lookup sha1
This function has been copied from the "patch_pos" function in "patch-ids.c" but an additional parameter has been added. The new parameter is a function pointer, that is used to access the sha1 of an element in the table. Signed-off-by: Christian Couder <chriscool@tuxfamily.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Christian Couder committed
Apr 4, 2009 at 22:59 UTC
96beef8c2efaab06f703991ed7802b8cef4c00e3
2 files changed
+108
sha1-lookup.c
+101
@@ -1,6 +1,107 @@
1
#include "cache.h"
2
#include "sha1-lookup.h"
3
4
+static uint32_t take2(const unsigned char *sha1)
5
+{
6
+ return ((sha1[0] << 8) | sha1[1]);
7
+}
8
+
9
+/*
10
+ * Conventional binary search loop looks like this:
11
+ *
12
+ * do {
13
+ * int mi = (lo + hi) / 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);
22
+ *
23
+ * The invariants are:
24
+ *
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).
29
+ *
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:
33
+ *
34
+ * - if it is a hit, we are happy.
35
+ *
36
+ * - if it is strictly higher than the target, we update hi with
37
+ * it.
38
+ *
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.
41
+ *
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.
47
+ */
48
+/*
49
+ * The table should contain "nr" elements.
50
+ * The sha1 of element i (between 0 and nr - 1) should be returned
51
+ * by "fn(i, table)".
52
+ */
53
+int sha1_pos(const unsigned char *sha1, void *table, size_t nr,
54
+ sha1_access_fn fn)
55
+{
56
+ size_t hi = nr;
57
+ size_t lo = 0;
58
+ size_t mi = 0;
59
+
60
+ if (!nr)
61
+ return -1;
62
+
63
+ if (nr != 1) {
64
+ size_t lov, hiv, miv, ofs;
65
+
66
+ for (ofs = 0; ofs < 18; ofs += 2) {
67
+ lov = take2(fn(0, table) + ofs);
68
+ hiv = take2(fn(nr - 1, table) + ofs);
69
+ miv = take2(sha1 + ofs);
70
+ if (miv < lov)
71
+ return -1;
72
+ if (hiv < miv)
73
+ return -1 - nr;
74
+ if (lov != hiv) {
75
+ /*
76
+ * At this point miv could be equal
77
+ * to hiv (but sha1 could still be higher);
78
+ * the invariant of (mi < hi) should be
79
+ * kept.
80
+ */
81
+ mi = (nr - 1) * (miv - lov) / (hiv - lov);
82
+ if (lo <= mi && mi < hi)
83
+ break;
84
+ die("oops");
85
+ }
86
+ }
87
+ if (18 <= ofs)
88
+ die("cannot happen -- lo and hi are identical");
89
+ }
90
+
91
+ do {
92
+ int cmp;
93
+ cmp = hashcmp(fn(mi, table), sha1);
94
+ if (!cmp)
95
+ return mi;
96
+ if (cmp > 0)
97
+ hi = mi;
98
+ else
99
+ lo = mi + 1;
100
+ mi = (hi + lo) / 2;
101
+ } while (lo < hi);
102
+ return -lo-1;
103
+}
104
+
105
/*
106
* Conventional binary search loop looks like this:
107
*
sha1-lookup.h
+7
@@ -1,6 +1,13 @@
1
#ifndef SHA1_LOOKUP_H
2
#define SHA1_LOOKUP_H
3
4
+typedef const unsigned char *sha1_access_fn(size_t index, void *table);
5
+
6
+extern int sha1_pos(const unsigned char *sha1,
7
+ void *table,
8
+ size_t nr,
9
+ sha1_access_fn fn);
10
+
11
extern int sha1_entry_pos(const void *table,
12
size_t elem_size,
13
size_t key_offset,