Fix issue #112: 'Dynamically calling a fixture causes a runtime error'

[ykuzma1]

This patch works by failing over to a ThreadPoolExecutor in cases where the expected event loop is already running. Notably this happens when calling request.getfixturevalue(argname) because it dynamically calls a fixture that needs to be setup on an event loop that's already running the asynchronous pytest tests.

[ykuzma1]

Sorry, I was a bit premature. I was hoping to avoid changing the behavior of the existing code as much as possible by trying to run setup() as before and only falling back to running within the Executor if a RuntimeException occurred. Unfortunately, this was running setup() multiple times (🤦‍♂️) causing issues with test setup and tear-down. This means that it has to run setup() in the Executor no matter what.

The only other option is detecting whether the fixture was called dynamically from within pytest_fixture_setup(), but I'm not aware of any way to do that.

[simonfagerholm]

@ykuzma1 How about adding a marker that can be identified from within the hooks?
So if you know that your testcase needs to call fixture dynamically you can mark that testcase

[asvetlov]

I guess the PR should be closed.
Running an async fixture setup phase in a thread pool violates the asyncio usage rules.
Thanks for the PR anyway!