Hello @kaxil , @kacpermuda found an issue with the backward compat baseoperator due to my PR which add's support for async callable in PythonOperator. This PR is a fix for that issue. Good catch BTW by @kacpermuda. Question, don't we run backward compat tests of providers with Airflow 2.x? I would have expected this to have failed tests.

Hello @kaxil , @kacpermuda found an issue with the backward compat baseoperator due to my PR which add's support for async callable in PythonOperator. This PR is a fix for that issue. Good catch BTW by @kacpermuda. Question, don't we run backward compat tests of providers with Airflow 2.x? I would have expected this to have failed tests.

Generally yes. We take all the tests we have for providers and run them against 2.11 -> https://github.com/apache/airflow/actions/runs/21096649515/job/60675322714 for example. But maybe there is no test to run this path or maybe it is disabled for Airflow 2 (some tests are disabled because they are not even supposed to work on Airflow 2) @kacpermuda -> @dabla. Maybe you can see if there was a test missing that would cover that path and add it ?

Note that we are not running "airflow 2" core tests in those tests - only provider tests. So my best guess is that none of the "standard" provider tests or "lineage" provider tests actually run the code that triggerred the issue and we should add it.

As Jarek said, probably not test for this or we did not test this path. I caught this when running actual Dags with inlets/outlets that use push_xcom. Running OL provider system tests on main is still on my todo list, and I think that would be a nice e2e coverage for such issues - as OpenLineage works so closely with Airflow via listener, the OL e2e tests have already helped found few bugs in airflow itself. Anyway, let me know if there is any way I can help in covering that specific test path @dabla and thanks for the quick fix PR.