refactor: Refactored __new__ magic method of BaseOperatorMeta to avoid bad mixing classic and decorated operators

[dabla]

After the discussion started by @potiuk on the dev list, I've started a new pull request in which I implemented a decorator which checks from where the execute method of an operator is being called. Maybe the check will be to restrictive, so I hope we will see the effect of this in the integration tests as the check is disabled when running unit tests, otherwise the execute method of an operator would never be able to run. I've added 2 unit tests, one case where it's being called directly and thus fails and another one where it's indirectly being called from a TaskInstance.

Any remarks are welcome.

^ Add meaningful description above
Read the Pull Request Guidelines for more information.
In case of fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
In case of a new dependency, check compliance with the ASF 3rd Party License Policy.
In case of backwards incompatible changes please leave a note in a newsfragment file, named {pr_number}.significant.rst or {issue_number}.significant.rst, in newsfragments.

[uranusjr]

Is it possible (desirable?) to raise an error when the operator is instantiated, instead of when execute is called?

[dabla]

Is it possible (desirable?) to raise an error when the operator is instantiated, instead of when execute is called?

I've tried that at first as that was also my first idea, but then the decorator will already be applied (or not) when the class get's loaded instead of when calling the executed method, which then makes it impossible to test because it's to late as I modify the unit_test_mode env var before each run of the test.

[Taragolis]

Also check failed Static Code Checks: https://github.com/apache/airflow/actions/runs/8172329572/job/22342633909

[dabla]

I've tried refactoring it with sentinel but now integration tests start failing probably because it's considering the operator being called outside of TaskInstance while this was not the case with traceback.

[dabla]

I've tried refactoring it with sentinel but now integration tests start failing probably because it's considering the operator being called outside of TaskInstance while this was not the case with traceback.

Ok found out why this is failing, it because now the test_mode flag is used from TaskInstance but in those failing tests it's not and it's calling the execute methode directly on operator as a unit test (celery).

[potiuk]

@ashb ?

[ashb]

Happy with the code now, I haven't caught up on the mailing list discussion though

[potiuk]

LGTM. If @ashb is ok with it, we can merge it now for 2.9.0. Having a warning is "good enough" for now. If we decide to do something more (for example allow such use to behave more "as expected") we can always add it later.