rjf@eagle.ukc.ac.uk (Robin Faichney) (07/29/88)
Sorry about this -- the proper place for this article would be the sharedx mailing list, but I've been trying unsuccessfully to get on to that for some time (mainly local problems, I think) and as I won't be at Xhibition for the sharedx meeting, I'd like to air what I have to say as soon as possible. I strongly believe that inter-program communications (IPC -- I don't think the overloading is confusing) are not, generally, the concern of any window management system, and that X should not be used to implement remote conferencing. Sharing window input and output, and handling multiple displays -- fine. These are user-interface issues, and X is part of the user-interface. Where IPC is all about sharing user-interface resources, by all means let it be mediated by the user-interface resource manager: in this case the X server. But where non-user-interface communications are concerned, that is part of the underlying application functionality, which (I'm sure we are all agreed) should be separated, as far as possible, from the user-interface. Why should X have the job of arranging things which have nothing (or no more than any other part of an interactive application's functionality) to do with user I/O? Merely because it already incorporates such funky IPC stuff? That's not much of a reason, but it's the only one I can think of. Here's the embarrassing bit: I have designed and implemented a system (dp) which arranges IPC with the minimum of effort from the application programmer, and which is fully networking, thanks in part to the MIT people and their X11.2 sample server and client code. We have a remote conferencing system (vconf), which is built upon dp, primarily as a test and demo for it. Vconf runs on Sun3s and VAXstations, under X10, X11 and Sunview, and it neither needs nor wants any Xtensions! Both dp and vconf are presently in use here, but it will be a little while before we attempt anything beyond the realms of /etc/hosts (for non-U*IX-types -- the local network) -- though the general concept at least seems fairly solid so far. We (the Kent Software Tools Group) are planning to release both dp and vconf, but no schedule has yet been set. I'm not allowed to mention what we have available right now, because that would be advertising ;-( -- but if anyone is interested in dp, etc, please contact myself in the first instance: Robin Faichney, Computing Lab, The University, Canterbury, Kent. CT2 7NF. UK. or rjf@ukc.ac.uk