diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/JavaTypeInference.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/JavaTypeInference.scala index 61c153c10e47c24834c6ea230496cc541b204c8b..2de066f99498d5c6d34eece5b123de165e7ba19a 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/JavaTypeInference.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/JavaTypeInference.scala @@ -334,7 +334,11 @@ object JavaTypeInference { */ def serializerFor(beanClass: Class[_]): CreateNamedStruct = { val inputObject = BoundReference(0, ObjectType(beanClass), nullable = true) - serializerFor(inputObject, TypeToken.of(beanClass)).asInstanceOf[CreateNamedStruct] + val nullSafeInput = AssertNotNull(inputObject, Seq("top level input bean")) + serializerFor(nullSafeInput, TypeToken.of(beanClass)) match { + case expressions.If(_, _, s: CreateNamedStruct) => s + case other => CreateNamedStruct(expressions.Literal("value") :: other :: Nil) + } } private def serializerFor(inputObject: Expression, typeToken: TypeToken[_]): Expression = { @@ -417,7 +421,7 @@ object JavaTypeInference { case other => val properties = getJavaBeanProperties(other) if (properties.length > 0) { - CreateNamedStruct(properties.flatMap { p => + val nonNullOutput = CreateNamedStruct(properties.flatMap { p => val fieldName = p.getName val fieldType = typeToken.method(p.getReadMethod).getReturnType val fieldValue = Invoke( @@ -426,6 +430,9 @@ object JavaTypeInference { inferExternalType(fieldType.getRawType)) expressions.Literal(fieldName) :: serializerFor(fieldValue, fieldType) :: Nil }) + + val nullOutput = expressions.Literal.create(null, nonNullOutput.dataType) + expressions.If(IsNull(inputObject), nullOutput, nonNullOutput) } else { throw new UnsupportedOperationException( s"Cannot infer type for class ${other.getName} because it is not bean-compliant") diff --git a/sql/core/src/test/java/test/org/apache/spark/sql/JavaDatasetSuite.java b/sql/core/src/test/java/test/org/apache/spark/sql/JavaDatasetSuite.java index 8304b728aa2389635e82991f2f36c4e6cfc91315..b25e3493c17b6783ba116e3eb3f708c5b52acbef 100644 --- a/sql/core/src/test/java/test/org/apache/spark/sql/JavaDatasetSuite.java +++ b/sql/core/src/test/java/test/org/apache/spark/sql/JavaDatasetSuite.java @@ -1305,4 +1305,28 @@ public class JavaDatasetSuite implements Serializable { spark.createDataset(data, Encoders.bean(NestedComplicatedJavaBean.class)); ds.collectAsList(); } + + @Test(expected = RuntimeException.class) + public void testNullInTopLevelBean() { + NestedSmallBean bean = new NestedSmallBean(); + // We cannot set null in top-level bean + spark.createDataset(Arrays.asList(bean, null), Encoders.bean(NestedSmallBean.class)); + } + + @Test + public void testSerializeNull() { + NestedSmallBean bean = new NestedSmallBean(); + Encoder<NestedSmallBean> encoder = Encoders.bean(NestedSmallBean.class); + List<NestedSmallBean> beans = Arrays.asList(bean); + Dataset<NestedSmallBean> ds1 = spark.createDataset(beans, encoder); + Assert.assertEquals(beans, ds1.collectAsList()); + Dataset<NestedSmallBean> ds2 = + ds1.map(new MapFunction<NestedSmallBean, NestedSmallBean>() { + @Override + public NestedSmallBean call(NestedSmallBean b) throws Exception { + return b; + } + }, encoder); + Assert.assertEquals(beans, ds2.collectAsList()); + } }