[ On Tue, June 23, 1998 at 09:50:12 (-0400), Perry E. Metzger wrote: ]
> Subject: Re: word processor that runs on NetBSD/i386? (FAQ?) 
>
> I like the idea of the plan 9 abstraction myself. I would prefer to
> get rid of the non-file namespaces and non-file access methods for the 
> kernel. Unfortunately, practicality interferes here.

Unfortunately one such interface that's unlikely to go away any time soon no matter
what wonderful feats of programming are achieved is ioctl(2).

Even more unfortunately the current NetBSD implementation of ioctl()
generally makes it completely and absolutely impossible for simple
backward compatabilty of these interfaces even when something as benign
as a new field is added at the end of a structure.  I appreciate the
correctness of this feature, but it came as an extreme surprise since I
was very accustomed to far less error checking with ioctl() interfaces.

For example I have as yet been unable to upgrade my current firewall
because I'm unable to run a -current kernel with ip-filter on a non
-current user-land.  Indeed even though I can build a -current kernel on
my non -current systems I'm unable to build the affected tools without
disrupting the build environment of these systems.  In this particular
case with ip-filter I can see that the change should be benign and
shouldn't affect any of the normal operations I need to perform, but
because the ioctl() data structure size changed *nothing* works.

This wouldn't normally be as much of a problem for people who can afford
to run separate build machines.  However even with a proper build
environment this kind of restriction is still quite annoying on slower
and older machines.

If only libkvm users were the only tools that were adversely affected by
changes to the kernel innards.....

-- 
							Greg A. Woods

+1 416 443-1734      VE3TCP      <gwoods@acm.org>      <robohack!woods>
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>