Aiohttp: Test Client Loop Error with Actual App

Long story short

The suggested testing pattern requires boilerplate duplication of app code and doesn't show testing the app directly.

https://docs.aiohttp.org/en/stable/testing.html#pytest

Expected behaviour

I should be able to pass my app module directly to test_client to test my app.

Actual behaviour

The docs show building the app from handlers on the fly before each test. Passing the real app directly to test_client works fine the first time, then fails with "RuntimeError: web.Application instance initialized with different loop".

Steps to reproduce

# handler.py

async def foo():
    return 'foo'

# app.py

from aiohttp import web
from .handler import foo

app = web.Application()
app.add_routes([web.get('/foo/', foo)])

# test.py

from .app import app

async def test_foo_status(test_client):
    response = await client.get('/foo/')
    assert response.status == 200

async def test_foo_body(test_client):
    response = await client.get('/foo/')
    body = await response.text()
    assert body == 'foo'

Your environment

• OS: Ubuntu 18.04
• Python: 3.6.5
• aiohttp: 3.3.2

Work around

I hacked around the issue for now by peeking at the source code and determining the cause of the error.

        if self._loop is not None and self._loop is not loop:
            raise RuntimeError(
                "web.Application instance initialized with different loop")

https://github.com/aio-libs/aiohttp/blob/e561eaa/aiohttp/web_app.py#L134

It is not immediately clear to my why this code is necessary. I could not find any context in the git history. Setting app._loop = None in my test cleanup method seems to avoid the problem without any negative side effects. Is this code actually necessary? Could this fix be integrated into test_client?