Re: Hurd on Minix3 or other kernels?

[ness]

Filip Brcic wrote:
> Дана Thursday 17 November 2005 02:43, B. Douglas Hilton је написао(ла):
>
> > I've been thinking about what I wrote above, and one kind of nice thing
> > is that it should be possible to bootstrap the Minix3/Hurd system by
> > compling from with a chrooted environment. I built a Gentoo system from
> > within a Debian xterm by doing this kind of trickery. For instance, you
> > could set up some build space, unpack your Hurdish binutils, and attempt
> > to compile them using the Minix compiler. Once they are built, use them
> > to attempt to compile glibc0 or other low-level things.
> >
> > Since Minix is bootable, has a clib, and a compiler, a bit of iterative
> > bootstrap compiling in a chrooted environment might be worth checking
> > out. At the moment all four of my primary partitions are filled, so I
> > can't install Minix3 to fool around with it, but perhaps with a bit of
> > carefull hard drive juggling I could free up a primary for this purpose.
> > If any of you know a way to install Minix3 into a secondary and boot it
> > with GRUB, then do please spill the beans because I have a couple nice
> > chunks of freespace in my extended partition.
>
>
> Why don't you try to run minix3 from some kind of virtual machine (qemu, 
> bochs, plex86, vmware, etc)? Or buy another disk (for 100€ one can get 160+ 
> GB disk and play around as much as one likes).
>
> Anyway, porting GNU/Hurd to Minix3 sounds like an interesting idea. If you can 
> get the Hurd to be kernel-independent (as long as the kernel is microkernel), 
> it would be very useful for Hurd/L4, Hurd/X2, Hurd/Darwin, or Hurd/<other>.
This is ilusoric, IMHO. When I came to the Hurd, I asked why this wasn't 
done, too. But now, as I'm a bit more familiar with micrkernels, I see 
the following: a microkernel has to export a minimal interface to be 
efficient. But this means, the interfaces might be extremely different. 
Have only a look at EROS and L4Ka::Pistachio, two recent microkernels. 
In L4 you have recoursive adress spaces constructed by sending map items 
via ipc. In EROS you have adress spaces constructed in user space using 
capabilities to pages. Don't mind Mach. What I want to say is: the 
level, where you can make the Hurd kernel independant is a really high 
one. Not to forget the planned design changes, of that at least some 
seem to be incompatible to me.
--
-ness-