added AsyncSingleThreadContext

[Arfey]

I have a problem with async database connections for Django TestCase.

The psycopg3 connection is not thread-safe. As a result, I need to create a new connection per thread. This is a problem for the TestCase coz it uses a transaction for each test case.

class TestCase(TransactionTestCase):
    async def _apost_teardown(self):
        for conn in async_connections.all(initialized_only=True):
            await conn.close()

    def _setup_and_call(self, result, debug=False):
        if iscoroutinefunction(testMethod):
            ############### convert test 
            setattr(self, self._testMethodName, async_to_sync(testMethod))

        if debug:
            super().debug()
        else:
            super().__call__(result)

       self._post_teardown()
       ############### convert teardown 
       async_to_sync(self._apost_teardown)()       

I can do something like that coz each async_to_sync function uses a new thread executor (and different connection as a result). To solve that, I try to implement something similar for ThreadSensitiveContext.

class TestCase(TransactionTestCase):
    def _setup_and_call(self, result, debug=False):
        with AsyncSingleThreadContext()
            if iscoroutinefunction(testMethod):
                ############### convert test 
                setattr(self, self._testMethodName, async_to_sync(testMethod))

            if debug:
                super().debug()
            else:
                super().__call__(result)

           self._post_teardown()
           ############### convert teardown 
           async_to_sync(self._apost_teardown)()       

[fcurella]

Hi! how can we help getting this reviewed?

[carltongibson]

Hey @fcurella. You can review it. Tell me it works 🙂

[fcurella]

@Arfey the code LGTM, but I think she should add some documentation in docs/ the README and the CHANGELOG

[carltongibson]

I'd forgotten about this 🤹‍♀️ I've been meaning to get to your PR @fcurella. Again 🤹‍♀️.

When I looked at this initially I thought "seems sensible".

Looking again (Friday afternoon) the description is a bit quick/loose. A comment along the lines of "yes, this looks correct to me because ..." would help me pick it up again.

Happy to get a release out with it if we're confident.

[fcurella]

Suggested change

	# They should not have run in the main thread, and on the same threads
	# They should have run in the main thread, and on the same threads

[fcurella]

I've verified that:

1. The new context manager picks the event loop's main thread if available, or creates a new thread.
2. Nested uses of the context manager keep using the thread selected by the outermost one.

[carltongibson]

Ok, great thanks. Let me look at it properly next week.

[carltongibson]

https://pypi.org/project/asgiref/3.10.0/

Thanks both!