Took 3.8 out of the allow failures list since we need to now support 3.8.

@sethmlarson We still have the start_tls() test failing on 3.8, like it did back in #423 - see my #423 (comment).

I spent 30mins trying to debug what it is that we're doing wrong, but I can't seem to find anything. Where did you find the info/resources about how to implement the transport switching (see #263)? Could it be there's something we're missing there?

Locally I'm receiving an asyncio error/warning:

ERROR    asyncio:base_events.py:1692 An open stream object is being garbage collected; call "stream.close()" explicitly.

which is emitted by this piece of code. This is a new warning in 3.8 that was contributed here: python/cpython#9201.

The warning is even emitted using:

import asyncio

async def main(host, port):
    stream_reader, stream_writer = await asyncio.wait_for(  # type: ignore
        asyncio.open_connection(host, port), 5
    )
    stream_writer.close()
    del stream_writer  # Simulate garbage collection

asyncio.run(main("localhost", 8000))

By adding await asyncio.sleep(0) just after stream_writer.close() I am able to remove the error; so I added that to our Stream.close() implementation to see what was left.

Well, the error is still there when running the start_tls() test — which makes me think we might not be cleaning up either the old stream_reader or stream_writer properly?

(I also tried the await asyncio.sleep(0) trick inside .start_tls(), i.e. just after we've called start_tls() and obtained a new transport. But the effect is that a ConnectionResetError is raised when we write the request during the test. So… 🤷‍♂)

Hi @jtmiclat, now that #484 I believe you can rebase from master and the build with 3.8 should pass. :)

Kicked another build off, let's see if we can merge now without a rebase. :)

@jtmiclat Thank you for this change :) 🎉