Re: Build while creating kernel-headers package

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

Manoj Srivastava <srivasta@ieee.org> wrote:
> Hi,
> 
>        It was remarked on IRC that a build with the new
> kernel-package does a full build even when compiling kernel-headers,
> and this behaviour was different from the kernel-package in Sid.
> 
>        Now, since make-kpkg only calls the top level makefile with a
> very limited set of targets, namely, one of the configure targets, a
> build, or an install target, one can figure (without delving into
> code or looking at logs, like I did) that all we used to do before
> was call make oldconfig or similar before creating a kernel headers
> package. 
> 
>        So, I did a build with a modified new kernel package, and did
> a debdiff on a kernel headers package  created after a full build,
> with another with just a make oldconfig; and I discovered that not
> doing a build creates a kernel header package missing things like
> include/linux/compile.h and Module.symvers.
> 
>        So, while it is true that the behaviour has changed, it is not
> a regression in kernel-package, as has been  stated, it is a bug
> fix. Full debdiff below.
> 
>        I wonder, if this deficiency was know, and the official
> headers packages had work arounds, why was  this not brought to the
> attention of the kernel-package maintainer? This deficiency has left
> the users of kernel-package down, and I am sorry about that.  I would
> have thought caring for end users would have been more important than
> turf wars.

Firstly, its not about turf wars. If you wish to attack it from that
angle, then I shall not participate in the discussion any further. 
The finger pointing situation is, frankly, rediculous.

The headers package has often been broken, so it wouldn't surprise me if
things are missing. The method that is used to create them involves a
combination of kernel-package and a postinst, which basically uses find.
This is available in SVN. And the results are examinable in the packages
available on debian.org. 

I would certainly welcome anyone looking into the headers packages and
finding deficiencies, and better still solutions. As I said earlier in
the week, though perhaps not on this channel, I am more than happy for
the common/flavour headers splitting that the the kernel-team's
packaging does moved into kernel-package. As I said above, its long
been a source of pain and frustration.

I encourage you to play with the linux-2.6 packages and explore
what in there is broken, how we could be using kernel-package better,
and what logic might be best moved into kernel-package. I particularly
encourage this in the area of the headers packages.

-- 
Horms