Github user mn-mikke commented on a diff in the pull request:
https://github.com/apache/spark/pull/21687#discussion_r204340969
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala
---
@@ -695,6 +695,41 @@ abstract class TernaryExpression extends Expression {
}
}
+/**
+ * A trait resolving nullable, containsNull, valueContainsNull flags of
the output date type.
+ * This logic is usually utilized by expressions combining data from
multiple child expressions
+ * of non-primitive types (e.g. [[CaseWhen]]).
+ */
+trait ComplexTypeMergingExpression extends Expression {
+
+ /**
+ * A collection of data types used for resolution the output type of the
expression. By default,
+ * data types of all child expressions. The collection must not be empty.
+ */
+ @transient
+ lazy val inputTypesForMerging: Seq[DataType] = children.map(_.dataType)
+
+ /**
+ * A method determining whether the input types are equal ignoring
nullable, containsNull and
+ * valueContainsNull flags and thus convenient for resolution of the
final data type.
+ */
+ def areInputTypesForMergingEqual: Boolean = {
+ inputTypesForMerging.length <= 1 || inputTypesForMerging.sliding(2,
1).forall {
+ case Seq(dt1, dt2) => dt1.sameType(dt2)
+ }
+ }
+
+ override def dataType: DataType = {
+ require(
+ inputTypesForMerging.nonEmpty,
+ "The collection of input data types must not be empty.")
+ require(
+ areInputTypesForMergingEqual,
+ "All input types must be the same except nullable, containsNull,
valueContainsNull flags.")
--- End diff --
```NullType``` is ```sameType``` equal only with itself. So it's up to
coercion rules to cast ```NullType``` to a common type of other children. All
the rules transitively utilize ```TypeCoercion.findTightestCommonType``` that
does that job. ( ```case (NullType, t1) => Some(t1)```, ```case (t1,
NullType) => Some(t1)```)
If I don't miss anything, this behavior was the same before this PR. I can
add more tests to cover this scenario if you wish. :-)
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
