gabriel rosenkoetter wrote:
> 
> On Wed, Jan 09, 2002 at 03:13:43PM +0000, David Laight wrote:
> > Ah - now I see what is 'wrong'...
> >
> > Surely it is inappropriate to build makefs with the 'host' compiler and
> > header files.
> 
> Well, only sort of. makefs wants to create file systems the kernel
> supports. But if it's going to be part of the toolchain, it must
> work without knowledge of the running system.

But since such a filesystem cannot be used until the new kernel is built
the toolchain shouldn't need makefs.  If we are going to populate the
new
filesystem we have to be running a kernel that knows about it - and
hence we
can (presumably) use the native makefs.
> 
> > So why is it being built, and what is it used for?
> 
> It's a handy utility to create any supported file system within a
> file. (A generalization of mkisofs, in the conceptual sense.) makefs
> itself isn't really the problem here, rather its something wrong in
> the way the Makefile is dealing with it when it's built as part of
> the (stand-alone, non-installation) toolchain.

But makefs won't populate the filesystem - so where does the toolchain
use it?
The only hope for it generating a valid (empty) filesystem is that the
on-disk
filesystem format hasn't changed  even though the structures used by the
device
driver (when accessing the filesystem) have.

I can't see a valid use for makefs until you are running with the new
kernel.
> 
> There's a thread in tech-toolchain (beginning with Message-ID:
> <20020109041513.J25417@grappa.eclipsed.net>, or
> http://mail-index.netbsd.org/tech-toolchain/2002/01/09/0000.html,
> though the responses haven't hit mail-index yet) about this that might
> interest those troubled by it.

At the moment 'current' doesn't build on a 1.5.2 base......  In theory
it is supposed to cross build on non-netbsd systems!

	David