RE: [boost] Re: sockets Interface for address class

From: Dale Peakall (dale.peakall_at_[hidden])
Date: 2002-03-01 09:37:35

> > Not all types of addresses are network addresses with port numbers,
> > service names etc. There is "unix domain" address type in unices, 
> and
> > the address itself is just a pathname somewhere in file tree. 
> Obviously,
> > such an address doesn't require any resolving, and still it is
> > absolutely usable address for socket binding etc. (if the socket 
> itself
> > created as "unix domain" type socket).
> 
> Yes, and in general we shouldn't make any assumptions about the 
> representation of an address, so each protocol (or protocol family) 
> would probably have it's own address class (which could be something 
> as simple as string -- say, for "UNIX domain" sockets or WinNT pipes).

I thought that we'd sort of decided that what we were writing here
was a TCP/IP library, not sockets in the more general sense.

If you're going to start going down the multi-protocol route, then 
things really start getting complex.

Things are complex enough as they are.  Do you try and support
asynchronous communications?  I think the design document made it
clear that this was a simple TCP library.

If you want to make it multi-protocol then I for one would like to 
see a full OSI stack as well (with RFC1006 - OSI over TCP as just 
one of the supported protocols)!

        - Dale.