Re: [hybi] An alternative design for the web socket handshake

[Greg Wilkins <[e-mail]>]

Maciej Stachowiak wrote:
> The main downside is that it would be impossible to do non-TLS WebSocket; 

It would not be impossible.

If websocket was allocated a dedicated port for non-TLS connections,
then that could be used at least for development and testing.

TLS would only need to be used for actual deployment and the nextprotoneg
extension does allow easy sharing of port 443.  However I'm not sure how
long it will take before nextprotoneg is well supported by TLS libraries?

Note that the possibility of using non-HTTP handshake suggests that even
if we start with HTTP upgrade, it would be worthwhile to not depend too
much on HTTP headers in the handshake.  Having something like the
control frames in Scott Fergusons proposal will work regardless of the
way the connection is made.


cheers