|
| 1 | +#include "cache.h" |
| 2 | +#include "sha1-lookup.h" |
| 3 | + |
| 4 | +/* |
| 5 | + * Conventional binary search loop looks like this: |
| 6 | + * |
| 7 | + * unsigned lo, hi; |
| 8 | + * do { |
| 9 | + * unsigned mi = (lo + hi) / 2; |
| 10 | + * int cmp = "entry pointed at by mi" minus "target"; |
| 11 | + * if (!cmp) |
| 12 | + * return (mi is the wanted one) |
| 13 | + * if (cmp > 0) |
| 14 | + * hi = mi; "mi is larger than target" |
| 15 | + * else |
| 16 | + * lo = mi+1; "mi is smaller than target" |
| 17 | + * } while (lo < hi); |
| 18 | + * |
| 19 | + * The invariants are: |
| 20 | + * |
| 21 | + * - When entering the loop, lo points at a slot that is never |
| 22 | + * above the target (it could be at the target), hi points at a |
| 23 | + * slot that is guaranteed to be above the target (it can never |
| 24 | + * be at the target). |
| 25 | + * |
| 26 | + * - We find a point 'mi' between lo and hi (mi could be the same |
| 27 | + * as lo, but never can be as same as hi), and check if it hits |
| 28 | + * the target. There are three cases: |
| 29 | + * |
| 30 | + * - if it is a hit, we are happy. |
| 31 | + * |
| 32 | + * - if it is strictly higher than the target, we set it to hi, |
| 33 | + * and repeat the search. |
| 34 | + * |
| 35 | + * - if it is strictly lower than the target, we update lo to |
| 36 | + * one slot after it, because we allow lo to be at the target. |
| 37 | + * |
| 38 | + * If the loop exits, there is no matching entry. |
| 39 | + * |
| 40 | + * When choosing 'mi', we do not have to take the "middle" but |
| 41 | + * anywhere in between lo and hi, as long as lo <= mi < hi is |
| 42 | + * satisfied. When we somehow know that the distance between the |
| 43 | + * target and lo is much shorter than the target and hi, we could |
| 44 | + * pick mi that is much closer to lo than the midway. |
| 45 | + * |
| 46 | + * Now, we can take advantage of the fact that SHA-1 is a good hash |
| 47 | + * function, and as long as there are enough entries in the table, we |
| 48 | + * can expect uniform distribution. An entry that begins with for |
| 49 | + * example "deadbeef..." is much likely to appear much later than in |
| 50 | + * the midway of the table. It can reasonably be expected to be near |
| 51 | + * 87% (222/256) from the top of the table. |
| 52 | + * |
| 53 | + * The table at "table" holds at least "nr" entries of "elem_size" |
| 54 | + * bytes each. Each entry has the SHA-1 key at "key_offset". The |
| 55 | + * table is sorted by the SHA-1 key of the entries. The caller wants |
| 56 | + * to find the entry with "key", and knows that the entry at "lo" is |
| 57 | + * not higher than the entry it is looking for, and that the entry at |
| 58 | + * "hi" is higher than the entry it is looking for. |
| 59 | + */ |
| 60 | +int sha1_entry_pos(const void *table, |
| 61 | + size_t elem_size, |
| 62 | + size_t key_offset, |
| 63 | + unsigned lo, unsigned hi, unsigned nr, |
| 64 | + const unsigned char *key) |
| 65 | +{ |
| 66 | + const unsigned char *base = table; |
| 67 | + const unsigned char *hi_key, *lo_key; |
| 68 | + unsigned ofs_0; |
| 69 | + static int debug_lookup = -1; |
| 70 | + |
| 71 | + if (debug_lookup < 0) |
| 72 | + debug_lookup = !!getenv("GIT_DEBUG_LOOKUP"); |
| 73 | + |
| 74 | + if (!nr || lo >= hi) |
| 75 | + return -1; |
| 76 | + |
| 77 | + if (nr == hi) |
| 78 | + hi_key = NULL; |
| 79 | + else |
| 80 | + hi_key = base + elem_size * hi + key_offset; |
| 81 | + lo_key = base + elem_size * lo + key_offset; |
| 82 | + |
| 83 | + ofs_0 = 0; |
| 84 | + do { |
| 85 | + int cmp; |
| 86 | + unsigned ofs, mi, range; |
| 87 | + unsigned lov, hiv, kyv; |
| 88 | + const unsigned char *mi_key; |
| 89 | + |
| 90 | + range = hi - lo; |
| 91 | + if (hi_key) { |
| 92 | + for (ofs = ofs_0; ofs < 20; ofs++) |
| 93 | + if (lo_key[ofs] != hi_key[ofs]) |
| 94 | + break; |
| 95 | + ofs_0 = ofs; |
| 96 | + /* |
| 97 | + * byte 0 thru (ofs-1) are the same between |
| 98 | + * lo and hi; ofs is the first byte that is |
| 99 | + * different. |
| 100 | + */ |
| 101 | + hiv = hi_key[ofs_0]; |
| 102 | + if (ofs_0 < 19) |
| 103 | + hiv = (hiv << 8) | hi_key[ofs_0+1]; |
| 104 | + } else { |
| 105 | + hiv = 256; |
| 106 | + if (ofs_0 < 19) |
| 107 | + hiv <<= 8; |
| 108 | + } |
| 109 | + lov = lo_key[ofs_0]; |
| 110 | + kyv = key[ofs_0]; |
| 111 | + if (ofs_0 < 19) { |
| 112 | + lov = (lov << 8) | lo_key[ofs_0+1]; |
| 113 | + kyv = (kyv << 8) | key[ofs_0+1]; |
| 114 | + } |
| 115 | + assert(lov < hiv); |
| 116 | + |
| 117 | + if (kyv < lov) |
| 118 | + return -1 - lo; |
| 119 | + if (hiv < kyv) |
| 120 | + return -1 - hi; |
| 121 | + |
| 122 | + if (kyv == lov && lov < hiv - 1) |
| 123 | + kyv++; |
| 124 | + else if (kyv == hiv - 1 && lov < kyv) |
| 125 | + kyv--; |
| 126 | + |
| 127 | + mi = (range - 1) * (kyv - lov) / (hiv - lov) + lo; |
| 128 | + |
| 129 | + if (debug_lookup) { |
| 130 | + printf("lo %u hi %u rg %u mi %u ", lo, hi, range, mi); |
| 131 | + printf("ofs %u lov %x, hiv %x, kyv %x\n", |
| 132 | + ofs_0, lov, hiv, kyv); |
| 133 | + } |
| 134 | + if (!(lo <= mi && mi < hi)) |
| 135 | + die("assertion failure lo %u mi %u hi %u %s", |
| 136 | + lo, mi, hi, sha1_to_hex(key)); |
| 137 | + |
| 138 | + mi_key = base + elem_size * mi + key_offset; |
| 139 | + cmp = memcmp(mi_key + ofs_0, key + ofs_0, 20 - ofs_0); |
| 140 | + if (!cmp) |
| 141 | + return mi; |
| 142 | + if (cmp > 0) { |
| 143 | + hi = mi; |
| 144 | + hi_key = mi_key; |
| 145 | + } |
| 146 | + else { |
| 147 | + lo = mi + 1; |
| 148 | + lo_key = mi_key + elem_size; |
| 149 | + } |
| 150 | + } while (lo < hi); |
| 151 | + return -lo-1; |
| 152 | +} |
0 commit comments