> I think the rationale at one point was that the kernel should
> provide a uniform interface to such things as mice, so that user
> programs do not have to worry about different mouse protocols.

It's not just user programs.  The average user generally has no idea
whether their mouse speaks the PS/2 mouse protocol or the Mouseman
protocol or what.  They can probably figure out whether it's a serial
or bus mouse if you tell them to check whether the connector in the
back of the machine is round or rectangular.

> For folks coming from Linux and/or FreeBSD (or other places, even)
> this seems to be a fairly constant source of confusion,

Probably because neither we nor XFree86 have done anything to the X
server or its documentation to reflect our approach.  So while it's
simpler for users in theory, they get lied to by the documentation
or setup tools (e.g. "man XF86Config").

(We don't, actually, get a whole lot of questions about this locally
at MIT/SIPB, since people using the local installation here are
directed to a simple five-question X setup script called "xconfig"
which understands the NetBSD approach.  So they get the simplicity
advantage, assuming they don't do anything dumb like read the manual.)