Re: #537: Remove segments (consensus call)

Because h2 is being designed as http only (as it is chartered to do), then
the only way that other semantics are going to be able to transit the h2
web is by making themselves look like http semantic!    Otherwise they will
be back at some 85% level because of intermediaries that don't accept
extensions and/or can only handle http semantic.

This is precisely the kind of protocol misuse the charter also asks us to

charter fail I think!


PS. if we fix HOL blocking, then http over h2 is a much more flexible
semantic for carrying websocket like semantic as things like long polling
will work much better.   But it will be very hard for a client to work out
the best transport to use as it does not really know if h1 or h2 will be
used or if it is better to try websocket.

On 16 July 2014 09:21, David Krauss <> wrote:

> What happened to WebSockets? Was a decision made off-list?
> Some means of flushing a stream is invaluable to interactive applications.
> A policy of one stream per request eats server (and intermediary)
> resources, and exhausts stream iDs.
> Don’t tell me that the traditional web isn’t interactive. Just because
> WebSockets aren’t yet widely deployed, doesn’t mean it isn’t the direction
> things are headed. HTTP/2 should be accelerating that process.
> On 2014–07–16, at 1:10 AM, Martin Thomson <>
> wrote:
> > On 14 July 2014 22:30, Mark Nottingham <> wrote:
> >> It sounds like we've converged on getting rid of END_SEGMENT. I've
> marked it as editor-ready.
> >
> > Since we're removing affordances for websockets, can I handle #557 at
> > the same time.  Not by adding METADATA, but by removing random,
> > semantic-free HEADERS frames.
> >

Greg Wilkins <> HTTP, SPDY, Websocket server and client that scales  advice and support for jetty and cometd.

Received on Tuesday, 15 July 2014 23:38:22 UTC