@aljoscha I agree with you. The current chain of dependencies does make refactoring difficult and not easy to abstract.
My question is that for different versions of kafka client, if the API has hardly changed (for example, kafka 0.11 and kafka 1.0), you are not in favor of flink-connector-kafka-1.0 inheriting the implementation of flink-connector-kafka-0.11? ? Or your idea is to extract some util methods (if the other connectors are independent of each other except the base module)? But these util methods may only work for both versions and not for all. If we don't take an inherited solution, it will result in a lot of duplicate code. In addition, I have done some common implementation abstraction for all kafka connector test code, but I don't know why the 0.9 version of the connector can't pass the test. I plan to fall back to the first commit, then split multiple commits to change it, and push each commit to the PR to verify that it doesn't break the existing tests and simplify the complexity of the positioning problem.What do you think of this? cc @pnowojski [ Full content available at: https://github.com/apache/flink/pull/6577 ] This message was relayed via gitbox.apache.org for [email protected]
