test_repository_tool: Add missing clear_all flag

[jku]

Without this the next create_*db() call will fail meaning
pytest test_repository_tool.py
fails on every test after this one.

Signed-off-by: Jussi Kukkonen [email protected]

[joshuagl]

Thanks for the patch @jku ! I am not familiar with pytest, so I'm not sure why this is required.

The setUp() method, which should be called at the start of each test, is already invoking clear_*db(clear_all=True). AIUI that should handle things not being correctly cleaned up in tearDown() (though maybe cleaning up correctly is the better way to do things, and we should remove those calls from setUp()?

The tests do work in our CI and if I invoke locally from within the tests subdirectory with python3 -m unittest -q test_repository_tool.

Should pytest work with standard library unittests? OOI why use pytest and not tox?

[jku]

Sorting in order of importance

The tests do work in our CI and if I invoke locally from within the tests subdirectory with python3 -m unittest -q test_repository_tool.

Should pytest work with standard library unittests?

I have no idea because I thought they were the same thing: they don't seem to be. So this is probably just me not knowing about python testing and not a bug.

why use pytest and not tox?

To the question "why not use tox": the tox setup seems a little broken to me as there's no way to run a single test because the default environment runs coverage... I've tried to write a new environment config for that but it becomes a little clunky. Calling "python3 -m unittest" seems fine to me: I just didn't know it

The setUp() method, which should be called at the start of each test, is already invoking clear_*db(clear_all=True). AIUI that should handle things not being correctly cleaned up in tearDown() (though maybe cleaning up correctly is the better way to do things, and we should remove those calls from setUp()?

This is only true for one of the classes I think: the others do not invoke the clear functions in setUp() at all -- this one was the only class that does not use clear_all out of all the classes so I figured that was the issue.

[jku]

Actually I think it is a real bug: if you rename TestTimestamp so it isn't the last test unittest runs, your command fails as well

[lukpueh]

To the question "why not use tox": the tox setup seems a little broken to me as there's no way to run a single test because the default environment runs coverage... I've tried to write a new environment config for that but it becomes a little clunky. Calling "python3 -m unittest" seems fine to me: I just didn't know it

FYI, you can run unittest-powerd modules/classes/tests directly like so:

python test_repository_tool.py # Runs all test cases (classes) in that module
python test_repository_tool.py TestRepository # Runs all tests (functions) in TestRepository test case
python test_repository_tool.py TestRepository.test_writeall # Runs test_writeall test only

Not sure if we can configure tox to allow this fine-grained control, or if that is actually necessary. On my box it would definitely take too long to create a fresh virtual environment and install everything with tox each time I want to run a specific test function. So I usually just run it directly in my dev virtual environment, which uses a recent Python version and has everything installed. I only run tox before I push.

[lukpueh]

@jku, do you think it would be helpful to have some of above info in our contributor testing docs?

[jku]

@jku, do you think it would be helpful to have some of above info in our contributor testing docs?

They seem obvious in hindsight ...but yes: I was following those instructions and did not figure out how to tweak them to run a single test case.

[jku]

wrote a better commit message (without the pytest confusion), also removed the superfluous cleanup Joshua noticed -- I think that's probably a workaround of a similar issue earlier in history.

[joshuagl]

Thank you for the fix, the discussion and the cleaned up commit message.