Github user lindblombr commented on a diff in the pull request:
https://github.com/apache/spark/pull/21847#discussion_r205912461
--- Diff:
external/avro/src/main/scala/org/apache/spark/sql/avro/AvroSerializer.scala ---
@@ -165,16 +183,112 @@ class AvroSerializer(rootCatalystType: DataType,
rootAvroType: Schema, nullable:
result
}
- private def resolveNullableType(avroType: Schema, nullable: Boolean):
Schema = {
- if (nullable) {
- // avro uses union to represent nullable type.
- val fields = avroType.getTypes.asScala
- assert(fields.length == 2)
- val actualType = fields.filter(_.getType != NULL)
- assert(actualType.length == 1)
- actualType.head
+ // Resolve an Avro union against a supplied DataType, i.e. a LongType
compared against
+ // a ["null", "long"] should return a schema of type Schema.Type.LONG
+ // This function also handles resolving a DataType against unions of 2
or more types, i.e.
+ // an IntType resolves against a ["int", "long", "null"] will correctly
return a schema of
+ // type Schema.Type.LONG
+ private def resolveNullableType(avroType: Schema, catalystType: DataType,
+ nullable: Boolean): Schema = {
+ (nullable, avroType.getType) match {
+ case (false, Type.UNION) | (true, Type.UNION) =>
+ // avro uses union to represent nullable type.
+ val fieldTypes = avroType.getTypes.asScala
+
+ // If we're nullable, we need to have at least two types. Cases
with more than two types
+ // are captured in test("read read-write, read-write w/ schema,
read") w/ test.avro input
+ assert(fieldTypes.length >= 2)
--- End diff --
I cleaned up this logic to be more clear.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
