Don't run unmarked tests in the reactor greenlet.

[tomprince]

I have a project using pytest-twisted, and I'd like to write some tests using hypothesis' stateful testing. Since the control-flow of stateful testing is handled by hypothesis, I use pytest_twisted.blockon in the tests to interact with twisted.

However, this doesn't work by default, as pytest-twisted forces every test function to run in the reactor greenlet. There is a comment in the pytest_pyfunc_call hook suggesting this change:

pytest-twisted/pytest_twisted.py

Lines 357 to 363 in cc03b11

	def pytest_pyfunc_call(pyfuncitem):
	"""Interface to async test call handler."""
	# TODO: only handle 'our' tests? what is the point of handling others?
	# well, because our interface allowed people to return deferreds
	# from arbitrary tests so we kinda have to keep this up for now
	_run_inline_callbacks(_async_pytest_pyfunc_call, pyfuncitem)
	return not None

This is perhaps a duplicate of #64.

[altendky]

Based on the comment it seems the tension is simply between moving forward to a sensible place where you explicitly request pytest-twisted handle's your tests and the desire to not break backwards compatibility... bah. Perhaps a sensible transitional tool would be an ability to explicitly mark a test to not be handled by pytest-twisted? Would you consider that a useful and reasonable middle ground to address this in the short term?

I'll go ahead and note that while I do have a little familiarity with hypothesis, I don't have any with their stateful testing. As such, my commentary here, at least for now, won't have any hope of offering any specific insight about what a good way to integrate that with pytest-twisted would be. Though, having to use pytest_twisted.blockon() strikes me as a thing that we would like to avoid as much as possible.