From: Darryl Green (Darryl.Green_at_[hidden])
Date: 2002-11-25 19:26:28
> From: Hugo Duncan [mailto:hugoduncan_at_[hidden]]
> On Sun, 24 Nov 2002 10:12:36 +0100, Pavol Droba <droba_at_[hidden]>
> > Is there an interest to support also non-TCP/IP based protocols like
> > IRDA/TP or raw sockets?
> I think this should be feasable, though I know nothing of IRDA/TP.
> Is it just a case of using the appropriate sockaddr_?? class
> and protocol constants? Or are there any requirements on the
> communciation once connection is established?
I think it is important to consider that file (including/especially
device/special file eg. pipes) I/O should fit into the same framework. I
don't see much of a problem (beyond the name "socket") with this if I
understand the proposed architecture, which looks very similar to both
ACE and a lighter weight and imho easier to use ACE "replacement" we
developed at work. The acceptor/connector and address classes deal with
the peculiarities of creating/opening a socket. Once you have a
read/write "socket" you have a *nix file descriptor for most practical
purposes. If this model can be implemented portably the ability to use
at least some special "files" may also deal with some of the other
issues being discussed such as inter-thread/process communication when
using the async model (of course one way to do this might be to use unix
domain sockets where supported).
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk