Support AF_INET6 address family

[Yolley]

Closes #313 .

As stated here https://docs.python.org/2/library/socket.html#socket.AF_INET6, "For AF_INET6 address family, a four-tuple (host, port, flowinfo, scopeid) is used", so there should be no possible problems with type conversion on row 12 of this PR. So I think, that there is no need for additional tests, already existing should be enough to cover this.

[tomchristie]

Gotcha. Test cases currently failing, when looking up the “socket” extra info - not looked into it beyond that yet.

[Yolley]

Gotcha. Test cases currently failing, when looking up the “socket” extra info - not looked into it beyond that yet.

It seems to me, that MockTransport in tests misses 'socket' attribute. Here is a documentation for BaseTransport (which is being mocked in tests, as i understand) https://docs.python.org/3/library/asyncio-protocol.html#asyncio.BaseTransport.get_extra_info - 'socket' is also an attribute among with sockname, peername and sslcontext . I guess, we need to fix MockTransport in tests. Should we do it in this pr or create another?

[tomchristie]

Fixing it in this PR would be fine. However we ought to be a bit cautious about which items might or might not be present with get_extra_info. If we can fallback gracefully when particular items are not included then we should do that.

[Yolley]

Fixing it in this PR would be fine. However we ought to be a bit cautious about which items might or might not be present with get_extra_info. If we can fallback gracefully when particular items are not included then we should do that.

Like, first try to get 'socket', then, if it is not present, just return peername/sockname?

[tomchristie]

@Yolley Exactly, yes. I'd also suggest leaving the existing layout of the get_local_addr and get_remote_addr functions in place so that the change has a minimal-as-possible footprint.

def get_local_addr(transport):
    info = transport.get_extra_info("socket").  # Only these lines changed 
    if info is not None:                        # Only these lines changed 
        return info.getsockname()               # Only these lines changed 
    info = transport.get_extra_info("sockname")
    if info is not None and isinstance(info, (list, tuple)) and len(info) == 2:
        return (str(info[0]), int(info[1]))
    return None

[Yolley]

@Yolley Exactly, yes. I'd also suggest leaving the existing layout of the get_local_addr and get_remote_addr functions in place so that the change has a minimal-as-possible footprint.

def get_local_addr(transport):
    info = transport.get_extra_info("socket").  # Only these lines changed 
    if info is not None:                        # Only these lines changed 
        return info.getsockname()               # Only these lines changed 
    info = transport.get_extra_info("sockname")
    if info is not None and isinstance(info, (list, tuple)) and len(info) == 2:
        return (str(info[0]), int(info[1]))
    return None

I made changes in this PR like you suggested, but we still need to make some changes in MockTransport, we need to change logic of method get_extra_info. In documentation https://docs.python.org/3/library/asyncio-protocol.html#asyncio.BaseTransport.get_extra_info it is stated, that get_extra_info should return default, if name is not present in transport. So, I guess we need (for example, here https://github.com/encode/uvicorn/blob/master/tests/protocols/test_http.py#L86) to do something like this

    def get_extra_info(self, key):
        return {
            "sockname": self.sockname,
            "peername": self.peername,
            "sslcontext": self.sslcontext,
        }.get(key)

[tomchristie]

we still need to make some changes in MockTransport, we need to change logic of method get_extra_info

Yup, your suggestion there looks correct to me.

[Yolley]

we still need to make some changes in MockTransport, we need to change logic of method get_extra_info

Yup, your suggestion there looks correct to me.

So, i fixed MockTransport, now it uses get on dictionary in get_extra_info. Also I added 2 new tests to test socket logic in utils. In travis I see, that it suggests some reformat for test_utils.py, but I don't understand, what needs to be reformatted exactly.

[tomchristie]

Running ./scripts/lint will apply the correct formatting to the codebase.

[Yolley]

Running ./scripts/lint will apply the correct formatting to the codebase.

Yep, it seems, that black follows pep8 line of length (which is questionable in 2019, but it is still an approved standart, so). Reformatted test_utils.py

[tomchristie]

Released as 0.8.3. Good stuff! 👍