Re: sockets Interface for address class

From: dmoore99atwork (dmoore_at_[hidden])
Date: 2002-03-01 13:20:56

--- In boost_at_y..., Iain.Hanson_at_u... wrote:
>      
> Ahhh. Now I think I understand where the confusion is comming from. 
The 
> requirements state that we need to restrict ourselves to the 
network and 
> transport layers and not go off into http and ftp.
> 
> I have no desire to go against this requirements but IP is not the 
only network 
> layer protocol that is supported by the C sockets API.
> 
> I would like to reproduce, as much as possible, of the genericity 
of the C 
> sockets API with type saftey and without some of the redunancy.
> 
> /ikh

Yes, but I think what alot of folks are suggesting is that:

1.  We shouldn't do anything blatant to *prevent* the interface from 
working with Unix domain sockets, named pipes, etc.
2.  BUT, focus on a high quality implementation for the IP family 
(TCP and UDP over IPv4 and IPv6) that adds the type safety, steers 
the user away from common mistakes
and:
3.  Hopefully presents an elegant, platform independent means of 
using non-blocking or asynchronous communications.

Personally, I'd like to see the combined expertise of this board 
brought to bear on 2 and especially 3 before we provide concrete 
implementations for every socket protocol and address type under the 
sun....

I think the ideas you've been presenting are a GREAT foundation for 
#1, and what you're hearing from others is an eagerness to tackle the 
next items on the list.  Surely most people hear "socket" and think 
TCP or UDP...

Thanks,
Dave