Unilateral Stream Creation

[jasnell]

The current draft (-02) current states: "Streams can be established and used unilaterally."

Is this really the case? Can a server unilaterally decide to start sending unsolicited stream frames to a client without an associated PUSH_PROMISE or HTTP Request?

[martinthomson]

At the framing layer, yes. That said, the HTTP usage doesn't define any
semantics for this, so it's not legal for the protocol.

On 24 April 2013 15:15, James M Snell notifications@github.com wrote:

The current draft (-02) current states: "Streams can be established and
used unilaterally."

Is this really the case? Can a server unilaterally decide to start sending
unsolicited stream frames to a client without an associated PUSH_PROMISE or
HTTP Request?

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/73
.

[jasnell]

Then we need to clarify this in the spec... if we are defining the Framing layer and HTTP usage separately from one another, then we can say that while streams in general can be established and used unilaterally by the client or server, in sessions established specifically for HTTP/2.0 communication, servers MUST NOT establish new streams without an associated PUSH_PROMISE, and that PUSH_PROMISES may only be sent in client initiated streams.

This might be something we need clarification on list for, however.

[martinthomson]

Marking this as a design issue. It seems that we've had some discussion regarding stream "creation" that are inconsistent with the model that I had assumed. A fair number of folks seem to be under the impression that streams have to start with a particular type of message. Until we've discussed the specific issue, I'm not confident that we agree on this.

[mnot]

Discussed at SF Interim; pending Layering TF proposal.