Add Appveyor config for testing on Windows

[Mortal]

This PR adds an Appveyor config. When aio-libs/aiosmtpd is created on https://ci.appveyor.com (by the FLUFL), the config will be automatically picked up and tests will start running whenever master is updated.

Currently coverage fails because this if-branch is always taken in tests:

def eof_received(self):
    log.info('%r EOF received', self.session.peer)
    self._handler_coroutine.cancel()
    if self.transport is not None:                        # pragma: nopy34
        return super().eof_received()

I can't figure out how to make a test where eof_received() is called when self.transport is None, but it seems that this is exercised when testing on Linux by the tests in TestStartTLS (discovered by adding raise Exception() at the end of eof_received()).

[Mortal]

You will want to go to https://ci.appveyor.com/notifications and disable email notifications after signing up for Appveyor.

[warsaw]

I've enabled AppVeyor for aiosmtpd. I guess you're saying I want to disable email notifications because they'll be so spammy?

[warsaw]

As for your first comment about eof_received(), yes it does happen in Linux, but I'm not sure if this isn't a bad bandaid. OTOH it looks like maybe it's not a problem for you any more? In any case, this is a great branch, thanks for fixing things on Windows.

(I wonder if we need to start moving all those *-coverage.ini files out of the top-level?)

[Mortal]

Yeah, I personally think it's too much with an email per failed build.

The eof_received() coverage problem still persists. It seems like a bad band-aid to me -- I couldn't exactly figure out under which circumstances self.transport would become None, so I would say it needs an explanation in the code at the very least.

I think the *-coverage.ini should be moved out of the top-level. I tried to figure out if I could avoid having six distinct files, but I don't know enough about coverage to say if that is possible.

[warsaw]

On May 07, 2017, at 01:00 PM, Mathias Rav wrote: The `eof_received()` coverage problem still persists. It seems like a bad band-aid to me -- I couldn't exactly figure out under which circumstances self.transport would become None, so I would say it needs an explanation in the code at the very least.

I'm seeing that with my minor formatting branch. I'm going to investigate why that conditional is needed on Linux, which I hope will allow me to eliminate it or at least be able to describe the problem and find a way to cover it under Windows.

I think the `*-coverage.ini` should be moved out of the top-level. I tried to figure out if I could avoid having six distinct files, but I don't know enough about `coverage` to say if that is possible.

Unfortunately I don't think there is, but I've been meaning to engage with upstream coverage team to see if there's a better solution (maybe even help hack one up - e.g. something similar to tox's generated environments).

[warsaw]

My suspicion is that the traceback on eof_received() is a bug in asyncio. It actually doesn't affect the outcome of the aiosmtpd tests, but it's annoying console output. I'm going to file the bug in upstream Python and we'll see where that leads, but for now I can live with it.

Oh, and here's where I make a plea to upstream coverage to handle the multiple coverage.ini files better.