src/sha1_lookup.c

   1 /*
   2  * This file is basically taken from git code.
   3  * This file is free software; you can redistribute it and/or modify
   4  * it under the terms of the GNU General Public License, version 2,
   5  * as published by the Free Software Foundation.
   6  *
   7  * In addition to the permissions in the GNU General Public License,
   8  * the authors give you unlimited permission to link the compiled
   9  * version of this file into combinations with other programs,
  10  * and to distribute those combinations without any restriction
  11  * coming from the use of this file.  (The General Public License
  12  * restrictions do apply in other respects; for example, they cover
  13  * modification of the file, and distribution when not linked into
  14  * a combined executable.)
  15  *
  16  * This file is distributed in the hope that it will be useful, but
  17  * WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU General Public License
  22  * along with this program; see the file COPYING.  If not, write to
  23  * the Free Software Foundation, 51 Franklin Street, Fifth Floor,
  24  * Boston, MA 02110-1301, USA.
  25  */
  26
  27 #include <stdio.h>
  28
  29 #include "sha1_lookup.h"
  30 #include "common.h"
  31
  32 /*
  33  * Conventional binary search loop looks like this:
  34  *
  35  *      unsigned lo, hi;
  36  *      do {
  37  *              unsigned mi = (lo + hi) / 2;
  38  *              int cmp = "entry pointed at by mi" minus "target";
  39  *              if (!cmp)
  40  *                      return (mi is the wanted one)
  41  *              if (cmp > 0)
  42  *                      hi = mi; "mi is larger than target"
  43  *              else
  44  *                      lo = mi+1; "mi is smaller than target"
  45  *      } while (lo < hi);
  46  *
  47  * The invariants are:
  48  *
  49  * - When entering the loop, lo points at a slot that is never
  50  *   above the target (it could be at the target), hi points at a
  51  *   slot that is guaranteed to be above the target (it can never
  52  *   be at the target).
  53  *
  54  * - We find a point 'mi' between lo and hi (mi could be the same
  55  *   as lo, but never can be as same as hi), and check if it hits
  56  *   the target.  There are three cases:
  57  *
  58  *    - if it is a hit, we are happy.
  59  *
  60  *    - if it is strictly higher than the target, we set it to hi,
  61  *      and repeat the search.
  62  *
  63  *    - if it is strictly lower than the target, we update lo to
  64  *      one slot after it, because we allow lo to be at the target.
  65  *
  66  *   If the loop exits, there is no matching entry.
  67  *
  68  * When choosing 'mi', we do not have to take the "middle" but
  69  * anywhere in between lo and hi, as long as lo <= mi < hi is
  70  * satisfied.  When we somehow know that the distance between the
  71  * target and lo is much shorter than the target and hi, we could
  72  * pick mi that is much closer to lo than the midway.
  73  *
  74  * Now, we can take advantage of the fact that SHA-1 is a good hash
  75  * function, and as long as there are enough entries in the table, we
  76  * can expect uniform distribution.  An entry that begins with for
  77  * example "deadbeef..." is much likely to appear much later than in
  78  * the midway of the table.  It can reasonably be expected to be near
  79  * 87% (222/256) from the top of the table.
  80  *
  81  * However, we do not want to pick "mi" too precisely.  If the entry at
  82  * the 87% in the above example turns out to be higher than the target
  83  * we are looking for, we would end up narrowing the search space down
  84  * only by 13%, instead of 50% we would get if we did a simple binary
  85  * search.  So we would want to hedge our bets by being less aggressive.
  86  *
  87  * The table at "table" holds at least "nr" entries of "elem_size"
  88  * bytes each.  Each entry has the SHA-1 key at "key_offset".  The
  89  * table is sorted by the SHA-1 key of the entries.  The caller wants
  90  * to find the entry with "key", and knows that the entry at "lo" is
  91  * not higher than the entry it is looking for, and that the entry at
  92  * "hi" is higher than the entry it is looking for.
  93  */
  94 int sha1_entry_pos(const void *table,
  95                    size_t elem_size,
  96                    size_t key_offset,
  97                    unsigned lo, unsigned hi, unsigned nr,
  98                    const unsigned char *key)
  99 {
 100         const unsigned char *base = (const unsigned char*)table;
 101         const unsigned char *hi_key, *lo_key;
 102         unsigned ofs_0;
 103
 104         if (!nr || lo >= hi)
 105                 return -1;
 106
 107         if (nr == hi)
 108                 hi_key = NULL;
 109         else
 110                 hi_key = base + elem_size * hi + key_offset;
 111         lo_key = base + elem_size * lo + key_offset;
 112
 113         ofs_0 = 0;
 114         do {
 115                 int cmp;
 116                 unsigned ofs, mi, range;
 117                 unsigned lov, hiv, kyv;
 118                 const unsigned char *mi_key;
 119
 120                 range = hi - lo;
 121                 if (hi_key) {
 122                         for (ofs = ofs_0; ofs < 20; ofs++)
 123                                 if (lo_key[ofs] != hi_key[ofs])
 124                                         break;
 125                         ofs_0 = ofs;
 126                         /*
 127                          * byte 0 thru (ofs-1) are the same between
 128                          * lo and hi; ofs is the first byte that is
 129                          * different.
 130                          */
 131                         hiv = hi_key[ofs_0];
 132                         if (ofs_0 < 19)
 133                                 hiv = (hiv << 8) | hi_key[ofs_0+1];
 134                 } else {
 135                         hiv = 256;
 136                         if (ofs_0 < 19)
 137                                 hiv <<= 8;
 138                 }
 139                 lov = lo_key[ofs_0];
 140                 kyv = key[ofs_0];
 141                 if (ofs_0 < 19) {
 142                         lov = (lov << 8) | lo_key[ofs_0+1];
 143                         kyv = (kyv << 8) | key[ofs_0+1];
 144                 }
 145                 assert(lov < hiv);
 146
 147                 if (kyv < lov)
 148                         return -1 - lo;
 149                 if (hiv < kyv)
 150                         return -1 - hi;
 151
 152                 /*
 153                  * Even if we know the target is much closer to 'hi'
 154                  * than 'lo', if we pick too precisely and overshoot
 155                  * (e.g. when we know 'mi' is closer to 'hi' than to
 156                  * 'lo', pick 'mi' that is higher than the target), we
 157                  * end up narrowing the search space by a smaller
 158                  * amount (i.e. the distance between 'mi' and 'hi')
 159                  * than what we would have (i.e. about half of 'lo'
 160                  * and 'hi').  Hedge our bets to pick 'mi' less
 161                  * aggressively, i.e. make 'mi' a bit closer to the
 162                  * middle than we would otherwise pick.
 163                  */
 164                 kyv = (kyv * 6 + lov + hiv) / 8;
 165                 if (lov < hiv - 1) {
 166                         if (kyv == lov)
 167                                 kyv++;
 168                         else if (kyv == hiv)
 169                                 kyv--;
 170                 }
 171                 mi = (range - 1) * (kyv - lov) / (hiv - lov) + lo;
 172
 173 #ifdef INDEX_DEBUG_LOOKUP
 174                 printf("lo %u hi %u rg %u mi %u ", lo, hi, range, mi);
 175                 printf("ofs %u lov %x, hiv %x, kyv %x\n",
 176                        ofs_0, lov, hiv, kyv);
 177 #endif
 178
 179                 if (!(lo <= mi && mi < hi)) {
 180                         return git__throw(GIT_ERROR, "Assertion failure. Binary search invariant is false");
 181                 }
 182
 183                 mi_key = base + elem_size * mi + key_offset;
 184                 cmp = memcmp(mi_key + ofs_0, key + ofs_0, 20 - ofs_0);
 185                 if (!cmp)
 186                         return mi;
 187                 if (cmp > 0) {
 188                         hi = mi;
 189                         hi_key = mi_key;
 190                 } else {
 191                         lo = mi + 1;
 192                         lo_key = mi_key + elem_size;
 193                 }
 194         } while (lo < hi);
 195         return -((int)lo)-1;
 196 }