Re: Sun, Java, Linux

Hasdi R Hashim (hasdi@engin.umich.edu)
Sat, 8 Jun 1996 12:04:20 -0400 (EDT)


On Sat, 8 Jun 1996, Eric Plante wrote:
> On Wed, 5 Jun 1996, Matt Hartley wrote:
> > Tell that to Sun.
>
> Actually, if you consider that .class files _are_ compiled (for the Java
> VM, that is), then it you could argue that putting a "java machine"
> emulator in the kernel would make sense... You could even consider it as
> another binary format (which I think is the generally accepted point of
> view)..

I think what Linus and his advocates want is if it can be done in
userspace, then do it. If you need to mess around with hardward, use a
setuid root server. If you really really can't, maybe you can get little
kernel support extension (hooks is it?) and bigger user-space part. If you
really really really really really can't and no living human being can
live without, Linus MIGHT consider putting it into the kernel.

Me, I much rather type 'java packageA.HelloWorld' thank you very much.
Java interpreter choice vary too much. You can have a byte-by-byte
interpreter much like what Sun initially uses or just-in-time compiler, ie
compiles everything and then executes it. Maybe another company might
release a JIT than is tighter and faster, another more secure. In other
words, out to the userland it goes.

Maybe what linux-kernel really needs is better support for userland device
drivers, something that allow scsi and ide, filesystem device drivers work
as well in userspace. might bloat up the kernel a bit but well justified
if you can remove a big portion of the kernel to userland without too much
loss in performance.

Regards

Hasdi