Send an empty payload for NO_STATUS_RCVD

[mhils]

x = wsproto.events.CloseConnection(wsproto.frame_protocol.CloseReason.NO_STATUS_RCVD)
ws.send(x)

currently emits a close frame with status code 1000 (NORMAL_CLOSURE) instead of not sending a payload. This PR fixes this.
NO_STATUS_RCVD should probably be more generally called NO_STATUS_CODE, but that would unnecessarily break compatibility.

[pgjones]

If I understand correctly the 1005 (NO_STATUS_RCVD) should not be sent, but rather should be substituted if there is no status code (on received close frames).

I wonder if instead of this change the function should it raise a ValueError if the code passed in is NO_STATUS_RCVD on the basis that it can already send empty frames (code=None) and trying to explicitly send this code is an odd thing to do.

[pgjones]

Could you change to reason is not None

[mhils]

wsproto currently generates CloseConnection events with an empty string as the payload.

wsproto/src/wsproto/frame_protocol.py

Line 529 in 38716a9

reason = data[2:].decode("utf-8")

Would you prefer if we change that to a None payload?

[mhils]

If I understand correctly the 1005 (NO_STATUS_RCVD) should not be sent, but rather should be substituted if there is no status code (on received close frames).

Yes. The current behavior is that if one tries to send a 1005, it is silently substituted with a 1000. My proposal is to change that to not sending a status code instead. It would definitely be wrong to send an actual status code of 1005 on the wire.

---

The problem at its core is that wsproto currently uses the special value NO_STATUS_RCVD when receiving frames without status code, but asks for a different special value (code=None) to send them. That feels inconsistent and is annoying when proxying messages for example.

At the moment the CloseConnection also does not declare code as optional, so strictly speaking you can't use the high-level API to send empty close frames:

wsproto/src/wsproto/events.py

Lines 181 to 182 in 38716a9

	code: int
	reason: Optional[str] = None

My proposal would be we consistently use one special value (None or NO_STATUS_RCVD), or at least accept 1005 for sending as well so that it doesn't need to be special-cased. I'd be fine with either case, but since the spec already proposes 1005 that seems like a reasonable choice. Does that make sense?

[pgjones]

I've merged manually with a slight change to the reason None checking - as reason=="" has no affect on the payload and I wanted to keep the diff to the code change only.

[mhils]

Awesome, thanks for getting back at this!

[pgjones]

The reason change did in the end make sense (to keep the autobahn tests passing), have re-added in d970644.