Re: An OS in Ada, why not RTEMS ?

[James Ross <rem.jr@rem.webross.com>]

On Sun, 28 Apr 2002 21:13:12 GMT, nickroberts@ukf.net (Nick Roberts)
wrote:

>It is not entirely an exercise in reinventing the wheel. It has certain
>features that I have not seen in any other microkernel (certain deadlock
>avoidance and locking features). I believe these features to be vital (for
>the purposes of a distributed general-purpose OS).

>RTEMS is only a real-time kernel (no doubt a very good one). But it has no
>security features, and misses many other features that (I believe) an AdaOS
>kernel must have.

As Marin has pointed out in a previous thread, AdaOS should
distinguish itself from other OS's out there.  This would keep the
project from falling into the category of re-inventing the wheel.  I
agree on these features you are planning and it will be very
interesting to see your design on it.

>>It seems to use free GNAT, you would need to support either
>>ELF, or PE right of the bat in your new OS and be willing to do some
>>hacking on the GNAT run time sources as well.

>I would be delighted if someone were to take on the work of modifying GNAT
>and/or GCC enough to make it usable to build all the components of an OS in
>Ada.

Unless some GNAT guru steps in and proves otherwise, I don't think it
will be possible to create an OS from scratch using the free GNAT
unmodified.  If this were not a hobby project, of course I would be
willing to purchase the Professional GNAT version. Certainly it is
possible to do it there :)

>However, one of my key personal objectives in starting the AdaOS project
>was to create not just a new microkernel, but a new Ada compiler which
>would specifically target this microkernel. This is ambition enough, so I
>feel I cannot devote any time to modifying GNAT at this stage.

I am certain that the Ada community here will welcome another Ada
compiler! I can sympathize with your decision to create a new Ada
compiler. It does make a lot of sense, since you will be able to
precisely control every aspect of the code generation and make it as
efficient on the AdaOS platform as you want.

>>I have come full circle in my thinking and believe it would be simpler
>>to start with say either FreeBSD or Linux and gradually "craft" it
>>into an "Ada OS".  You would not need to hack the free GNAT toolset.
>>Of the many advantages to this approach, one of them would be that
>>even if the goal is never reached, you might end up with some really
>>cool add-ons and/or extensions to an existing OS. Isn't this basically
>>what Apple has done with OS X?
>
>Maybe the AdaOS project should go this way, or some other alternative
>route. It would have to do so without me, but with a membership of nearly
>50 now (I think), that surely wouldn't be a problem. I'm going to start
>discussing things like this with the membership soon.

Well, it is just an idea.  And I don't mean to change the path the
AdaOS project is on.  Ultimately, I like the idea of from scratch
best.  But if that is not happening, then perhaps the course does need
to be re-thought / re-planned. 

If I sit down and I am going to write an OS, I think "OK what do I
need"� I need  Keyboard, COM Ports, Network Cards, TCP/IP, Console,
Floppy, IDE, UDMA, SCSI, Sound Blaster, Parallel Ports, Graphic Cards,
USB, etc� etc� etc� that is what I mean about re-inventing the wheel.
And that's just the hardware stuff / protocols / etc � how about all
the GNU development stuff?.  In some form, you need the tools
represented there.  

If someone were to take all the ideas discussed of what an AdaOS
could/should be and implement them under an AdaOS source tree
(thunking and isolating the calls to the underlying OS so those
portions can be replaced in the future) then any program written using
it would be guaranteed to work on whatever future version of the
kernel / OS would be.  Anyway� it is just a thought!
JR