Re: Ada Microkernel?

["B. Douglas Hilton" <doug.hilton@engineer.com>]

Ok, let me mull this over for awhile amd figure out what to do.
It seems to me like building a microkernel with Ada should be
a rather pleasant task. Except I want to make it plug and play for
Debian/Hurd, and I really want to use OSKit for my hardware
drivers.
    This implies a strong linkage between C OSKit drivers and
mostly alien Ada code ( as far as the driver world is concerned. )
    Still, this is a great project. If OSKit-Mach could be redone in
Ada, and still provide mach.h and MIG, then the Debian/Hurd could
actually be ported to an Ada Mach-comaptible microkernel.
    Currently, Roland McGrath is the Hurd equivalent of Linus, and
it is exceedingly difficult to hack the GNUMach or OSKit-Mach
sources. Basically nobody except he really understands the new
Hurd microkernel.

I would like to change this. The Hurd is supposed to be the OS
for the rest of us. I think Hurd's kernel should be written in some
better language ( such as Ada ) so that it is easier to modify in
the future.

Have you tried Hurd yet? A 1Gb partition will get you going!
Download Hurd at http://www.debian/org/ports/hurd

Unfortunately, GNATS has not yet been ported to Hurd, but I
can attest to the platform's viability as I can boot Hurd and run
X11 and it really does work.

- Doug


"chris.danx" wrote:

> You don't need to have system level calls specifically for C applications, you
> have system level calls for all applications (not just C ones).  There's nothing
> special about C applications that means you have to have a special format for
> them.  If i were you I'd make the API Ada styled (strongly typed and well
> defined!  None of this "words are all you need" that you see in C based OS
> api's.
>
> The message passing systems used in microkernels (just one way of communication
> possible) seem to work well.  I've decided to take the standard message passing
> system(mailboxes) and try something that i've been told has been done in QNX and
> Beos.  The idea is like pattern matching from functional languages and a bit
> like dynamic typing.  It goes like this.
>
> Module A sends a message to B to do X.
>
> X can have more than one form, dependant on the types passed in.  You check the
> types of the items passed in (each item must have a type descriptor accompanying
> it) and X gives control to the appropriate routine.  If no routine is found you
> get an error (haven't really got a good idea about how to handle this,
> technically it should never happen but that doesn't mean it will never happen).
> You might want to consider something like this in an OS if your doing one.
>
> You could extend it to the API if you liked too.  (which is why i mentioned it).
>
> Chris Campbell