Re: [hybi] Websocket success rates and TLS extension.

[Greg Wilkins <[e-mail]>]

Julian Reschke wrote:
> On 17.04.2010 10:20, Greg Wilkins wrote:
>> ...
>> The other idea I had on a similar RTT would be to allow the
>> 101 response for an upgrade to contain a body.
>>
>> This a browser could send
>>
>>     GET /somethingUseful/andThenUpgrade
>>     Upgrade: WebSocket
>>
>> and the response would return a useful entity, thus not wasting
>> a RT, and then upgrade after the response body.
>>
>> Unfortunately I think this goes beyond the charter of
>> the current HTTPbis working group.
>> ...
> 
> 1xx messages only come with headers, right.
> 
> But then, can't you just put the payload into a new response header?

Indeed!  it could even be the very first frame sent out over
the websocket connection.

So if the user was a SPDY/BWTP like sub protocol over
websocket, then the conversation could go something like:


>  GET /somethingUseful
   Upgrade: Websocket

<                  HTTP/1.1 101 Switching Protocols

<                  0x00 frame containing HTTP response
                        and something useful 0xff

>  0x00 frame containing next HTTP request 0xFF
>  0x00 frame containing next next HTTP request 0xFF
<                  0x00 next HTTP response 0xFF
<                  0x00 next next HTTP response 0xFF

etc.

That solves the non TLS RTT issue pretty well I think!

cheers