Explicitly suggest FIN or RST_STREAM

[martinduke]

For issue #1878. I was apparently the only one confused, but since this is different from HTTP/2 it may be good to make it crystal clear.

[RyanTheOptimist]

Is it worth commenting that the receiver of the RST_STREAM frame MUST eventually send a FIN or a RST_STREAM in order for the stream to close completely?

[marten-seemann]

@RyanatGoogle That applies to every stream, not just streams that a RST_STREAM was received for, doesn't it?

[RyanTheOptimist]

@marten-seemann Yes, definitely. But I understood that the point of this CL was to help clarify the QUIC RST_STREAM behavior which is different from HTTP/2. Since the proposed text says, SHOULD, I wondered if some reader might not realize that eventually they MUST. But I'm happy with this PR as is if nobody else is concerned :)

[kazuho]

I agree with @RyanatGoogle that the proposed text (quoted below) is confusing.

It will often be the case that upon receipt of a RST_STREAM on a bidirectional stream, an endpoint will choose to stop sending data to the peer on that stream. If so, it SHOULD send a FIN or RST_STREAM for that stream to close the peer's receive stream state.

My view is that the text can be interpreted as allowing an endpoint to stop sending data on that stream without sending FIN or RST_STREAM.

I think that it would be beneficial to either avoid use of SHOULD (by for example replacing it to "needs to"), or change it to MUST.

[martinthomson]

Merged manually.