xkrogen commented on code in PR #36506:
URL: https://github.com/apache/spark/pull/36506#discussion_r948268086
##########
connector/avro/src/main/scala/org/apache/spark/sql/avro/AvroSerializer.scala:
##########
@@ -218,6 +218,17 @@ private[sql] class AvroSerializer(
val numFields = st.length
(getter, ordinal) => structConverter(getter.getStruct(ordinal,
numFields))
+ case (st: StructType, UNION) =>
+ val unionConvertor = newComplexUnionConverter(st, avroType,
catalystPath, avroPath)
+ val numFields = st.length
+ (getter, ordinal) => unionConvertor(getter.getStruct(ordinal,
numFields))
+
+ case (DoubleType, UNION) if nonNullUnionTypes(avroType) == Set(FLOAT,
DOUBLE) =>
+ (getter, ordinal) => getter.getDouble(ordinal)
+
+ case (LongType, UNION) if nonNullUnionTypes(avroType) == Set(INT, LONG)
=>
+ (getter, ordinal) => getter.getLong(ordinal)
Review Comment:
This does imply some loss of information if you were to do a round-trip from
Avro to SQL to Avro, since all records written out would have double/long
values even if the input was a float/int.
The representation chosen on the read path is inherently lossy, since we
discard the information about which union branch the datum originated from, so
I don't think there's anything we can do here to avoid this behavior. Just
wanted to point it out.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
