Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/3330#discussion_r101775264
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/codegen/CodeGenerator.scala
---
@@ -1463,21 +1465,23 @@ class CodeGenerator(
*/
def addReusableFunction(function: UserDefinedFunction): String = {
val classQualifier = function.getClass.getCanonicalName
- val fieldTerm = s"function_${classQualifier.replace('.', '$')}"
+ val functionSerializedData = serialize(function)
+ val fieldTerm =
+ s"""
+ |function_${classQualifier.replace('.',
'$')}_${DigestUtils.md5Hex(functionSerializedData)}
--- End diff --
I think encoding the serialized object as String is a good idea. I like the
implementation.
Regarding the name collisions. I did a test with two identical scalar
functions and everything worked well because `reusableMemberStatements` is a
hash set that deduplicates the terms.
Btw. what else do we need to do for scalar UDFs? They are injected with the
same method and hence should be serialized as well. Isn't the only missing
thing to add more tests? If yes, we could do that in this PR.
What do you think @sunjincheng121 and @wuchong?
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
