> Well, I've already determined (at least for me) that it's so hairy to allow
> a system to boot on old /dev nodes.  I'd love to be proven wrong, but after
> two days of looking at the screen and oodles of printed kernel source, I
> have decided that it is much easier for kernel hackers and end users to
> upgrade /dev just before rebooting into a new kernel.

I can't count the number of times that I've had to "back out" the
installation of a new OS version because the new version was too
buggy to use.   I've also occasionally found it useful to boot
an "old" kernel on a "new" filesystem, even if only to fsck and dump
the disks (if the new kernel can't do this without crashing -- as was
the case when NetBSD-sun3 did "everything dumps core" -- you need
to run the old kernel, or even the vendor kernel, to recover.
And there's not always a spare boot device around.)

Either of these is a lot easier to do if the "new" OS is compatible 
with the "old" /dev entries. 

So I consider it very valuable to retain the ability to use existing
/dev entries, and not require incompatible changes when upgrading.

Keith