What should GNU run on (was Re: what kinds of things . . .)

[Kemp at DOCKMASTER.NCSC.MIL]

Wolfgang Rupprecht writes:
 > There are much more "open" platforms than Suns.
 >
 > A 386 PC clone has a well defined *and* documented register layouts.
 > [ . . . ]
 >
 > Wouldn't it be ironic to spend a lot of effort to write a free kernel
 > for a proprietary hardware base?

Yea, wasn't Berkeley foolish for spending so much effort on writing a
kernel for a proprietary architecture like the VAX?  :-)

I don't have any proprietary knowledge about Sun's MMU, but in order to
write code for my WhizBang-2000 Sparcstation clone, I don't have to.
The whole reason Sun is licensing the Sparc architecture is to
*encourage* clones, to provide a large enough market for shrink-wrap
software vendors like Lotus to be interested.  You can't have
shrink-wrap software without an Application Binary Interface which is
"well defined *and* documented".

The kernal (GNU or otherwise) isolates the hardware from the
application, and applications that muck with registers, screen memory,
etc directly (like much PC software) are poor examples of software
engineering.  Sun itself has at least 4 Kernel Virtual Memory
architectures - sun3 for the 68020, sun3x for the 68030, sun4 for the
4/100 & 200 series, and sun4c for the Sparcstation-1.  The code that
depends on the MMU is isolated in the kvm libraries.  Presumably
Solbourne, Prisma, Toshiba, the Taiwanese, Koreans, and other clone
makers will have a similar arrangement.  To claim that a particular MMU
hardware design is a "standard" that software should know about is to
constrain future development.  I think an "opaque" interface as provided
by the ABI is more in line with today's state of the art.

Besides, I would much rather have a Sparcstation on my desk than a PC!

    Dave Kemp <Kemp at dockmaster.ncsc.mil>