hash-lookup.c

   1 #define USE_THE_REPOSITORY_VARIABLE
   2
   3 #include "git-compat-util.h"
   4 #include "hash.h"
   5 #include "hash-lookup.h"
   6 #include "read-cache-ll.h"
   7
   8 static uint32_t take2(const struct object_id *oid, size_t ofs)
   9 {
  10         return ((oid->hash[ofs] << 8) | oid->hash[ofs + 1]);
  11 }
  12
  13 /*
  14  * Conventional binary search loop looks like this:
  15  *
  16  *      do {
  17  *              int mi = lo + (hi - lo) / 2;
  18  *              int cmp = "entry pointed at by mi" minus "target";
  19  *              if (!cmp)
  20  *                      return (mi is the wanted one)
  21  *              if (cmp > 0)
  22  *                      hi = mi; "mi is larger than target"
  23  *              else
  24  *                      lo = mi+1; "mi is smaller than target"
  25  *      } while (lo < hi);
  26  *
  27  * The invariants are:
  28  *
  29  * - When entering the loop, lo points at a slot that is never
  30  *   above the target (it could be at the target), hi points at a
  31  *   slot that is guaranteed to be above the target (it can never
  32  *   be at the target).
  33  *
  34  * - We find a point 'mi' between lo and hi (mi could be the same
  35  *   as lo, but never can be the same as hi), and check if it hits
  36  *   the target.  There are three cases:
  37  *
  38  *    - if it is a hit, we are happy.
  39  *
  40  *    - if it is strictly higher than the target, we update hi with
  41  *      it.
  42  *
  43  *    - if it is strictly lower than the target, we update lo to be
  44  *      one slot after it, because we allow lo to be at the target.
  45  *
  46  * When choosing 'mi', we do not have to take the "middle" but
  47  * anywhere in between lo and hi, as long as lo <= mi < hi is
  48  * satisfied.  When we somehow know that the distance between the
  49  * target and lo is much shorter than the target and hi, we could
  50  * pick mi that is much closer to lo than the midway.
  51  */
  52 /*
  53  * The table should contain "nr" elements.
  54  * The oid of element i (between 0 and nr - 1) should be returned
  55  * by "fn(i, table)".
  56  */
  57 int oid_pos(const struct object_id *oid, const void *table, size_t nr,
  58             oid_access_fn fn)
  59 {
  60         size_t hi = nr;
  61         size_t lo = 0;
  62         size_t mi = 0;
  63
  64         if (!nr)
  65                 return -1;
  66
  67         if (nr != 1) {
  68                 size_t lov, hiv, miv, ofs;
  69
  70                 for (ofs = 0; ofs < the_hash_algo->rawsz - 2; ofs += 2) {
  71                         lov = take2(fn(0, table), ofs);
  72                         hiv = take2(fn(nr - 1, table), ofs);
  73                         miv = take2(oid, ofs);
  74                         if (miv < lov)
  75                                 return -1;
  76                         if (hiv < miv)
  77                                 return index_pos_to_insert_pos(nr);
  78                         if (lov != hiv) {
  79                                 /*
  80                                  * At this point miv could be equal
  81                                  * to hiv (but hash could still be higher);
  82                                  * the invariant of (mi < hi) should be
  83                                  * kept.
  84                                  */
  85                                 mi = (nr - 1) * (miv - lov) / (hiv - lov);
  86                                 if (lo <= mi && mi < hi)
  87                                         break;
  88                                 BUG("assertion failed in binary search");
  89                         }
  90                 }
  91         }
  92
  93         do {
  94                 int cmp;
  95                 cmp = oidcmp(fn(mi, table), oid);
  96                 if (!cmp)
  97                         return mi;
  98                 if (cmp > 0)
  99                         hi = mi;
 100                 else
 101                         lo = mi + 1;
 102                 mi = lo + (hi - lo) / 2;
 103         } while (lo < hi);
 104         return index_pos_to_insert_pos(lo);
 105 }
 106
 107 int bsearch_hash(const unsigned char *hash, const uint32_t *fanout_nbo,
 108                  const unsigned char *table, size_t stride, uint32_t *result)
 109 {
 110         uint32_t hi, lo;
 111
 112         hi = ntohl(fanout_nbo[*hash]);
 113         lo = ((*hash == 0x0) ? 0 : ntohl(fanout_nbo[*hash - 1]));
 114
 115         while (lo < hi) {
 116                 unsigned mi = lo + (hi - lo) / 2;
 117                 int cmp = hashcmp(table + mi * stride, hash,
 118                                   the_repository->hash_algo);
 119
 120                 if (!cmp) {
 121                         if (result)
 122                                 *result = mi;
 123                         return 1;
 124                 }
 125                 if (cmp > 0)
 126                         hi = mi;
 127                 else
 128                         lo = mi + 1;
 129         }
 130
 131         if (result)
 132                 *result = lo;
 133         return 0;
 134 }