stats: fix estimation in between row count (#5135) (#5140)

statistics/histogram.go

@@ -283,7 +283,7 @@ func (hg *Histogram) betweenRowCount(sc *variable.StatementContext, a, b types.D
 		return 0, errors.Trace(err)
 	}
 	if lessCountA >= lessCountB {
-		return hg.inBucketBetweenCount(), nil
+		return hg.totalRowCount() / float64(hg.NDV), nil
 	}
 	return lessCountB - lessCountA, nil
 }
@@ -295,15 +295,6 @@ func (hg *Histogram) totalRowCount() float64 {
 	return float64(hg.Buckets[len(hg.Buckets)-1].Count)
 }
 
-func (hg *Histogram) bucketRowCount() float64 {
-	return hg.totalRowCount() / float64(len(hg.Buckets))
-}
-
-func (hg *Histogram) inBucketBetweenCount() float64 {
-	// TODO: Make this estimation more accurate using uniform spread assumption.
-	return hg.bucketRowCount()/3 + 1
-}
-
 func (hg *Histogram) lowerBound(sc *variable.StatementContext, target types.Datum) (index int, match bool, err error) {
 	index = sort.Search(len(hg.Buckets), func(i int) bool {
 		cmp, err1 := hg.Buckets[i].UpperBound.CompareDatum(sc, &target)

statistics/selectivity_test.go

@@ -155,7 +155,7 @@ func (s *testSelectivitySuite) TestSelectivity(c *C) {
 		},
 		{
 			exprs:       "a > 1 and b < 2 and c > 3 and d < 4 and e > 5",
-			selectivity: 0.00123287439,
+			selectivity: 0.00352249826,
 		},
 	}
 	for _, tt := range tests {