Github user wuchong commented on a diff in the pull request:
https://github.com/apache/flink/pull/4105#discussion_r122613823
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/api/TableEnvironment.scala
---
@@ -358,20 +358,27 @@ abstract class TableEnvironment(val config:
TableConfig) {
* Registers an [[AggregateFunction]] under a unique name. Replaces
already existing
* user-defined functions under this name.
*/
- private[flink] def registerAggregateFunctionInternal[T: TypeInformation,
ACC](
+ private[flink] def registerAggregateFunctionInternal[T: TypeInformation,
ACC: TypeInformation](
name: String, function: AggregateFunction[T, ACC]): Unit = {
// check if class not Scala object
checkNotSingleton(function.getClass)
// check if class could be instantiated
checkForInstantiation(function.getClass)
- val typeInfo: TypeInformation[_] = implicitly[TypeInformation[T]]
+ val resultTypeInfo: TypeInformation[_] = implicitly[TypeInformation[T]]
+ val accTypeInfo: TypeInformation[_] = implicitly[TypeInformation[ACC]]
// register in Table API
functionCatalog.registerFunction(name, function.getClass)
// register in SQL API
- val sqlFunctions = createAggregateSqlFunction(name, function,
typeInfo, typeFactory)
+ val sqlFunctions = createAggregateSqlFunction(
--- End diff --
Hi @fhueske , I think the ACC type extraction is similar to `T` type of
`AggreateFunction[T, ACC]` and `T` type of `TableFunction[T]`. Currently, they
are both extracted and add to the internal representation of UDAGGs and UDTFs.
To keep the consistent, I followed the same way to add the acc type. If we
decide to adopt the approach that wrapping the registered functions, it would
be better to make the other type extractions follow this approach.
What do you think ?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
