Re: regarding "HURD"

[olafBuddenhagen]

Hi,

On Mon, Nov 13, 2006 at 08:17:30PM -0500, B Douglas Hilton wrote:

> >>One thing that the group *does* agree about -- and most notably, the
> >>architects of the original Hurd system agree about this -- is that
> >>the Mach kernel has failed us.
> >>    
> >When and where did the architects of the original Hurd state that?
> >I'm not around very long yet, so maybe I just missed it; but I really
> >never heard of that so far...
> >  
> The problem with the Mach kernel is manifold. Here are some of the
> basic problems:
[...]
> The sad fact is that Mach is pretty much incomprehensible and
> unmaintainable. [...] Like a rusted out old classic car, it would be
> cool if we could restore it, but currently none of us has the time or
> means to undertake such a vast project.

I very much doubt "restoring" it would take more effort than creating a
completely new system from scratch. The problem is rather that it's less
rewarding. Even in a good shape, it will still be just an old car --
while most people here are more interested in playing with batmobiles.

> 3. Impossible driver support. Mach has most of its drivers built into
> the microkernel rather than loaded as modules. Because of this the
> hardware support is limited to ancient hardware drivers that were
> created by the original team of university researchers and the later
> corporate workers who were paid to develop them. The attempt to meld
> Mach with the OS-Kit failed when the Hurd developers decided that it
> would be easier to start from scratch than keep patching an
> obsolescent (mega)microkernel.

This part is totally wrong. The motivation for Hurd/L4 and now ngHurd
has nothing to do with device drivers.

Gnumach uses (ancient) Linux drivers. The reason they are so bad is that
nobody takes the trouble to update the glue code to use newer drivers.
Sure, user space drivers could offer many advantages -- but the driver
situation in the current Hurd has very little to do with that.

Anyways, the shortcomings of Mach are (today) widely known and
acknowledged. But that's beside the point. I didn't talk at all about
whether Mach has failed us or not. (Though the claim that everyone here
agrees it has, is evidently wrong. I know there are a number of people
here, who -- while interested in improvements possible with modern
kernels -- consider the existing implementation on Mach perfectly viable
for the time being.)

My question was very specifically about the claim regarding the
architects of the original Hurd system. I've never seen them say
anything like that Mach has failed us. So I'm asking for pointers to
some information on that matter that must have escaped me -- or
alternatively an acknowledgement that the claim is simply wrong. I don't
think that's too much to ask?

-antrik-