In article <20001221120046.A12984@theory.cs.uni-bonn.de> you write:
>But why do we need to split ports? The result is that we have to build
>the distribution N times, while 90% of that code is shared. Where 90% is:
>
>- the MI kernel code.
>- most of the CPU specific kernel code

Erm, a couple of points:

- The distribution is much more than the kernel.  It should be possible to
build everything but the kernel once for all ARM systems.  To me, this is
independent of how many ports we have -- if the build system can't do this,
it needs fixing.

- Each kernel is built from scratch anyway, since they each have a separate
directory under sys/arch/arm32/compile.  Currently, the kernel build process
doesn't allow for sharing objects between separate kernels.  Thus, we
already have the same code re-compiled for each kernel.  Separating the
ports loses us nothing.

>Whoever wants to split ports even more should device a release building
>process that allows reusing the 90% objects already build on, say, some
>Strongarm build slave for building the, say, RiscPC kernel and install medi=
>a.

I don't see that this wins you very much.  I think we can assume that people
building systems for release (as opposed to those building for a single
system of their own) will have a reasonable amount of CPU grunt at their
disposal.  My build system (160 MHz PPC603e) takes probably a few tens of
minutes to compile a kernel (it's fast enough that I've never timed it),
which is nothing compared to the time taken to build the entire system.

-- 
Ben Harris                                                   <bjh21@netbsd.org>
Portmaster, NetBSD/arm26               <URL:http://www.netbsd.org/Ports/arm26/>