RE: [boost] Note about a socket/network library

From: Jon Kalb (jonkalb_at_[hidden])
Date: 2001-08-02 13:36:38

> -----Original Message-----
> From: Beman Dawes [mailto:bdawes_at_[hidden]]
> Sent: Wednesday, August 01, 2001 2:30 PM
> To: boost_at_[hidden]; boost_at_[hidden]
> Subject: RE: [boost] Note about a socket/network library
> 
> At 01:16 PM 8/1/2001, Jon Kalb wrote:
> 
>  >Apple's <X networking interface is called OpenTransport and it is
based
>  >on XTI (Streams).
> 
> Thanks for the information.
> 
> XTI is documented in Steven's Unix "Networking Programming" book
mentioned
> in the requirements document.

> Later he says
> interoperability between XTI and sockets is complete, as it is
determined
> by the TCP or UDP protocols, and not the API.  "The API we use to
write
> either the client or server makes no difference."  So it sounds like a
> Boost library stands a reasonable chance of being implementable using
XTI.

I think you may be misinterpreting what he is saying here. What he means
is that the on-the-wire result of either implementation is the same and
clients written with one API can interoperate with servers written with
the other API. That is like saying that a client written in SmallTalk
could work with a server written in Java. I don't think it follows that
"a Boost [socket] library stands a reasonable chance of being
implementable using XTI."

On the other hand, Iain says:
> One of the reasons I was calling this the network library is that I
had
> in mind the possibility of an XTI/TLI implementation. Possibly using a
> templatized bridge pattern.

My guess would be that worrying about XTI/TLI APIs would be a needless
complication for a socket library, but if Iain has some insight into a
design that can be implemented to support both API's without adding too
much complexity or compromising performance or portability, I say "power
to 'im."