Re: Bug#278729: kernel-image-2.6-k7: kernel-image-2.6*-k7* and 2.6*-686* should

[Horms <horms@verge.net.au>]

Sven Luther <sven.luther@wanadoo.fr> wrote:
> On Tue, Nov 08, 2005 at 12:53:47PM -0600, Manoj Srivastava wrote:
>> On Wed, 9 Nov 2005 00:22:23 +0900, Horms  <horms@debian.org> said: 
>> 
>> > Don't we need to handle the control file ourselves because we make
>> > multiple make-kpkg invocations which all come together to form a
>> > single source package?
>> 
>>         I am planning on constructing the control file from bits even
>>  in the default install so that the uml and xen images do not show up
>>  in the control file but are never built, and we get all the "package
>>  exists in control file but not in changes" warnings.  Perhaps there
>>  can be a set of mini-control files concatenated together into a
>>  control file before it is read by Make?

I'm still not entirely sure that I understand how this would fit into
what linux-2.6 is doing. Perhaps if I breifly explain what it does 
you can fill in the gaps.

First up there are templates in templates/ and defitions the
architectures, sub-architectures and flavours there are in arch/. These
definitions are growing to include information about dependancies,
kernel header directories and stuff like that. arch/ also includes
configure fragments.

Before dpkg-buildpackage is run, a script is run which produces
a control file that contains all the packages for all architectures,
this is built using the information in templates/ and arch/. Simiarly,
rules.gen, which rules drives to produce the packages on each
architecture is produced. Its combination of rules.gen and control which
allows architectures, sub architectures and flavours to be added and
removed in a flexible manner, just by manipulating arch/

At run time, some additional mangling occurs to produce the .config
files based on the fragments in arch/ In a nuthsell kernel trees
are unpacked, the appropriate .config is produced and placed in the tree
and make-kpkg is called.

I should note, that I didn't actually write the system, and I'm still
learning how to drive it. Waldi and Dilinger know a lot more about it,
perhaps they can fill in some of my gaps.

> Manoj, we already do this, see the templates dir and bin/gencontrol.py. I told
> you that earlier already.

Is there really any need to be so aggressive?

Yes there is logic in there, and yes its coming along quite nicely.
But to be honest, the more logic we can push back to kernel-package,
the less we have to worry about. And I'm all about worrying less.

-- 
Horms