[...]
> >
> > 3. Use separate build tree approach and ship single kernel sources
> that point
> > to different kernel build trees that are distributed as separate packages.
> > (link established by kheader) That saves us overhead of having sources
> > n x times. It almost certainly requires Makefile fiddling but not very
> large.
> > It may fail for some non-conformant Makefiles (but then anything can fail)
> 
> <disclaimer type="kernel 2.6 newbie">
> Since the binary kernel packages ship with vmlinux and binary modules,
> can they not be linked into the kernel-source-$flavour package, so that
> you need kernel-source-base and kernel-source-$flavour (which requires
> kernel-$flavour), thus avoiding the need to include two copies of many
> files? kernel-source would be a virtual provides in kernel-source-$flavour.
> 
> This assumes either modpost can use compressed modules, or we ship
> uncompressed modules in kernel binary packages (or some similar solution).
> 
> Of course, if this is what you meant, just ignore me.
> </disclaimer>
> 

no you are right it is exactly what I meant in next paragraph 4.

This requires possibly non-trivial changes in makefiles (at least non-trivial
with my level of experience with gmake). Besides there is quite nasty issue -
how do we know to use kernel binaries and not in-tree modules? it is new
can of worms.

> 
> >
> > 4. Teach modpost to extract module versions from /boot/vmlinuz and
> /lib/modules.
> > Any takers? :)
> >
> > So far two most attractive approaches are
> >
> > - disable modversions. Frankly speaking, it was meant to allow people
> to reuse
> > modules across different kernel versions - as it stands now I have
> never seen
> > that module from one release could be loaded on another. It seems to
> be more
> > trouble than it is worth.
> 
> You don't maintain binary kernel module packages I think? If it does
> work, it would make life easier ...
> 

just that I understand. You think it makes sense to go without modversions or not?
I personally do not like removing it (it gives extra safety); but if it the
only option so far I would better turn it off and let people test kernel. 

> >
> > - take approach 3. We need to teach users how to use 2.6 kbuild
> anyway, there
> > is quite a bit of jump, so we may teach them how to use separate build
> tree
> > at the same time.
> 
> Option 4 looks the best long-term solution (and useful to others too).
> 

if you manage to convince kbuild maintainers to supply makefile part I promise
to add compressed modules support to modpost. It is not as hard (on the verge
of being trivial :)


-andrey

Reply via email to