extended unidirectional streams can be closed or reset

[kazuho]

closes #2331.

[LPardue]

I think this is almost there. There is already some text recommending RESET_STREAM logic for push streams, which are not extensions (see section 4.2.4).

Perhaps we can generalise the statement?

Also typo a minor typo "s/torelate/tolerate"

[kazuho]

@LPardue Do you have a specific suggestion re how we can generalize? My assumption has been that we cannot generalize, because control stream and QPACK streams can never be closed or reset.

And thank you for catching the typo. Fixed that.

[LPardue]

Something like lose the first sentence and say instead "Some unidirectional stream types are allowed to be closed or reset during a connection. A receiver MUST tolerate unidirectional stream close or reset prior to reception of the type byte."

Not sure my version is any better...

[kazuho]

@LPardue Thank you for the clarification. I agree that we should refrain from having a text that might seen as to imply that a push stream cannot be reset prior to sending the type byte.

I've changed the approach to clarify the general rule (that the streams can be reset or closed) then name out the exceptions. Does that look better?

[LPardue]

Lgtm, thanks!

[MikeBishop]

Looks good.