support decimal arithmetic precision loss mode

velox/docs/functions/spark/config.rst

@@ -0,0 +1,21 @@
+========================
+SparkRegistration Configuration properties
+========================
+
+struct SparkRegistrationConfig
+---------------------
+.. list-table::
+   :widths: 20 10 10 70
+   :header-rows: 1
+
+   * - Property Name
+     - Type
+     - Default Value
+     - Description
+   * - allowPrecisionLoss
+     - bool
+     - true
+     - When true, establishing the result type of an arithmetic operation according to Hive behavior and SQL ANSI 2011 specification, i.e.
+       rounding the decimal part of the result if an exact representation is not
+       possible. Otherwise, NULL is returned when the actual result cannot be represented with the calculated decimal type. Now we support add,
+       subtract, multiply and divide operations.

velox/docs/functions/spark/decimal.rst

@@ -27,14 +27,35 @@ Multiplication
 
 Division
 --------
-
+If the function is registered with ``allowPrecisionLoss`` being true as default:
 ::
 
     p = p1 - s1 + s2 + max(6, s1 + p2 + 1)
     s = max(6, s1 + p2 + 1)
 
+If the function is registered with ``allowPrecisionLoss`` being false:
+::
+
+    intDig = min(38, p1 - s1 + s2);
+    decDig = min(38, max(6, s1 + p2 + 1));
+
+If ``indDig + decDig`` is more than 38:
+::
+
+    p = 38
+    s = decDig - (intDig + decDig - 38) / 2 - 1
+
+Otherwise:
+::
+
+    p = intDig + decDig
+    s = decDig
+
+Decimal Precision and Scale Adjustment
+<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
+
 For above arithmetic operators, when the precision of result exceeds 38,
-caps p at 38 and reduces the scale, in order to prevent the truncation of
+caps p at 38 and reduces the scale when allowing precision loss, in order to prevent the truncation of
 the integer part of the decimals. Below formula illustrates how the result
 precision and scale are adjusted.
 
@@ -43,6 +64,8 @@ precision and scale are adjusted.
     precision = 38
     scale = max(38 - (p - s), min(s, 6))
 
+Caps p and s at 38 when not allowing precision loss.
+
 Users experience runtime errors when the actual result cannot be represented
 with the calculated decimal type.
 

velox/docs/spark_functions.rst

@@ -11,6 +11,7 @@ The semantics of Spark functions match Spark 3.5 with ANSI OFF.
     functions/spark/bitwise
     functions/spark/decimal
     functions/spark/comparison
+    functions/spark/config
     functions/spark/string
     functions/spark/datetime
     functions/spark/array