On 17 October 2011 19:09, Remy Maucherat <rmaucher_at_redhat.com> wrote:
> > * Byte streams are not sufficient for full
> > implementation/configuration of protocols. You need access to the
> > socket API for controlling such this as half closes, linger setting
> > and a wide range of other tuning parameters available on that API.
>
> I don't see how this is going to happen, unless the support is built it.
>
I agree this is very difficult to achieve. Hence with my proposal, I didn't
try to.
It may be sufficient to allow container implementations to override
application ones, and it will be the container implementations that perform
such full implementations. Byte streams might be good enough for
applications to provide back up implementations, although close semantics
would have to be well defined etc.
> Overall, maybe application protocols are not realistic ...
>
My fear is that they may not be useful or realistic. We have only 1
example, websockets. I think it is easier for most containers to implement
support for websockets than it is for them to implement support for
extensible arbitrary protocols implemented in application space.
cheers