Re: [hybi] Multiple connections serialization and proxies

[Jamie Lokier <[e-mail]>]

Greg Wilkins wrote:
> 
>    On 22 July 2010 08:28, Ian Hickson <[1][e-mail]> wrote:
> 
>      There's no lack of extensibility and negotiation, both are built
>      into the
>      protocol. Future versions of the protocol can simply define new
>      option
>      fields that the client can include to announce support and that the
>      server
>      can include to announce opt-in. For example, the client could say:
>      Â  Sec-WebSocket-Compression: gzip bzip2
>      ...and the server might then respond with:
>      Â  Sec-WebSocket-Compression: gzip
> 
>    Ian,
>    That will not work, because old versions of websocket are constrained
>    by the current standard to reject connections with unknown
>    headers.  So if a future version introduced such a header, it
>    would immediately make new client unable to interoperate with old
>    servers.
>    If we do not allow arbitrary headers, then we will severely limit our
>    options for upgrading the protocol in the future.

Ian, +1 to What Greg said.  It sounds like maybe the current spec
doesn't behave as you intended?

The handshake must permit future clients to interop with the first
widely deployed websocket servers (whatever version of the standard).
That's why "wiggle room" for negotation has to be clearly present in
the first version that gets significant deployment.

It doesn't have to be completely arbitrary headers; it just needs to
be *something* that all compliant servers ignore and future servers
can recognise.

-- Jamie