fixed decimal scales calc, updated the tests

src/AggregateFunctions/AggregateFunctionAvg.h

@@ -33,24 +33,26 @@ struct AvgFraction
 
     /// Allow division by zero as sometimes we need to return NaN.
     /// Invoked only is either Numerator or Denominator are Decimal.
-    Float64 NO_SANITIZE_UNDEFINED divideIfAnyDecimal(UInt32 scale) const
+    Float64 NO_SANITIZE_UNDEFINED divideIfAnyDecimal(UInt32 num_scale, UInt32 denom_scale) const
     {
         if constexpr (IsDecimalNumber<Numerator> && IsDecimalNumber<Denominator>)
         {
+            const UInt32 result_scale = std::max(num_scale, denom_scale);
+
             if constexpr (std::is_same_v<Numerator, Decimal256> && std::is_same_v<Denominator, Decimal128>)
                 ///Special case as Decimal256 / Decimal128 = compile error (as Decimal128 is not parametrized by a wide
                 ///int), but an __int128 instead
                 return DecimalUtils::convertTo<Float64>(
-                    numerator / (denominator.template convertTo<Decimal256>()), scale);
+                    numerator / (denominator.template convertTo<Decimal256>()), result_scale);
             else
-                return DecimalUtils::convertTo<Float64>(numerator / denominator, scale);
+                return DecimalUtils::convertTo<Float64>(numerator / denominator, result_scale);
         }
 
         /// Numerator is always casted to Float64 to divide correctly if the denominator is not Float64.
         Float64 num_converted;
 
         if constexpr (IsDecimalNumber<Numerator>)
-            num_converted = DecimalUtils::convertTo<Float64>(numerator, scale);
+            num_converted = DecimalUtils::convertTo<Float64>(numerator, num_scale);
         else
             num_converted = static_cast<Float64>(numerator); /// all other types, including extended integral.
 
@@ -58,7 +60,7 @@ struct AvgFraction
             Float64, Denominator> denom_converted;
 
         if constexpr (IsDecimalNumber<Denominator>)
-            denom_converted = DecimalUtils::convertTo<Float64>(denominator, scale);
+            denom_converted = DecimalUtils::convertTo<Float64>(denominator, denom_scale);
         else if constexpr (DecimalOrExtendedInt<Denominator>)
             /// no way to divide Float64 and extended integral type without an explicit cast.
             denom_converted = static_cast<Float64>(denominator);
@@ -90,8 +92,9 @@ public:
     using Fraction = AvgFraction<Numerator, Denominator>;
     using Base = IAggregateFunctionDataHelper<Fraction, Derived>;
 
-    explicit AggregateFunctionAvgBase(const DataTypes & argument_types_, UInt32 scale_ = 0)
+    explicit AggregateFunctionAvgBase(const DataTypes & argument_types_,
-        : Base(argument_types_, {}), scale(scale_) {}
+        UInt32 num_scale_ = 0, UInt32 denom_scale_ = 0)
+        : Base(argument_types_, {}), num_scale(num_scale_), denom_scale(denom_scale_) {}
 
     DataTypePtr getReturnType() const final { return std::make_shared<DataTypeNumber<Float64>>(); }
 
@@ -124,12 +127,14 @@ public:
     void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         if constexpr (IsDecimalNumber<Numerator> || IsDecimalNumber<Denominator>)
-            static_cast<ColumnVector<Float64> &>(to).getData().push_back(this->data(place).divideIfAnyDecimal(scale));
+            static_cast<ColumnVector<Float64> &>(to).getData().push_back(
+                this->data(place).divideIfAnyDecimal(num_scale, denom_scale));
         else
             static_cast<ColumnVector<Float64> &>(to).getData().push_back(this->data(place).divide());
     }
 private:
-    UInt32 scale;
+    UInt32 num_scale;
+    UInt32 denom_scale;
 };
 
 template <class T>

src/AggregateFunctions/AggregateFunctionAvgWeighted.cpp

@@ -83,11 +83,14 @@ AggregateFunctionPtr createAggregateFunctionAvgWeighted(const std::string & name
     if (left_decimal && right_decimal)
         ptr.reset(create(*data_type, *data_type_weight,
             argument_types,
-            getDecimalScale((sizeof(*data_type) > sizeof(*data_type_weight)) ? *data_type : *data_type_weight)));
+            getDecimalScale(*data_type), getDecimalScale(*data_type_weight)));
     else if (left_decimal)
-        ptr.reset(create(*data_type, *data_type_weight, argument_types, getDecimalScale(*data_type)));
+        ptr.reset(create(*data_type, *data_type_weight, argument_types,
+            getDecimalScale(*data_type)));
     else if (right_decimal)
-        ptr.reset(create(*data_type, *data_type_weight, argument_types, getDecimalScale(*data_type_weight)));
+        ptr.reset(create(*data_type, *data_type_weight, argument_types,
+            // numerator is not decimal, so its scale is 0
+            0, getDecimalScale(*data_type_weight)));
     else
         ptr.reset(create(*data_type, *data_type_weight, argument_types));
 

tests/performance/avg_weighted.xml

@@ -20,6 +20,7 @@
         INSERT INTO perf_avg(num)
         SELECT toUInt64(UserID / (WatchID + 1)  * 1000000)
         FROM hits_100m_single
+        LIMIT 50000000
     </fill_query>
 
     <query>SELECT avg(num)                  FROM perf_avg FORMAT Null</query>

tests/queries/0_stateless/01035_avg_weighted_long.reference

@@ -1,5 +1,7 @@
 2.3333333333333335
 nan
+1.0
+1.0
 8
 nan
 8

tests/queries/0_stateless/01035_avg_weighted_long.sh

@@ -5,6 +5,8 @@ CUR_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 
 ${CLICKHOUSE_CLIENT} --query="SELECT avgWeighted(x, weight) FROM (SELECT t.1 AS x, t.2 AS weight FROM (SELECT arrayJoin([(1, 5), (2, 4), (3, 3), (4, 2), (5, 1)]) AS t));"
 ${CLICKHOUSE_CLIENT} --query="SELECT avgWeighted(x, weight) FROM (SELECT t.1 AS x, t.2 AS weight FROM (SELECT arrayJoin([(1, 0), (2, 0), (3, 0), (4, 0), (5, 0)]) AS t));"
+${CLICKHOUSE_CLIENT} --query="SELECT avgWeighted(x, y) FROM (select toDecimal256(1, 0) x, toDecimal256(1, 1) y);"
+${CLICKHOUSE_CLIENT} --query="SELECT avgWeighted(x, y) FROM (select toDecimal32(1, 0) x, toDecimal256(1, 1) y);"
 
 types=("Int8" "Int16" "Int32" "Int64" "UInt8" "UInt16" "UInt32" "UInt64" "Float32" "Float64")