Re: 2.6.14 kernel udebs and d-i

[Sven Luther <sven.luther@wanadoo.fr>]

On Tue, Nov 15, 2005 at 05:17:42PM -0500, Joey Hess wrote:
> Thought I'd document the process of updating d-i's kernel udebs from
> 2.6.12 to 2.6.14, since there's been various speculation about how
> automatable this is.

Just a little addition from a porter's perspective. I uploaded the powerpc
.udebs yesterday, and had to do :

  1) removed the now gone tcic module from pcmcia-modules.
  2) removed the nls_base module from fs-common-modules.
  3) removed both unix and af_packet from socket-modules, removed the
  socket-modules .udeb altogether. These have been builtin since 2.6.12 i
  hear, altough maybe not for powerpc.
  4) added the mii module to nic-shared-modules, as it was included by
  dependency in both nic-modules and nic-extra-modules and the build
  complained.
  5) added a dependency from pcmcia-modules to firmware-modules as
  pcmcia-modules pulled in firmware_class and thus there where duplicated
  modules again.

That's it, and only nls_base and tcic are even remotely porter related, and
they where the most trivial fixes, and could easily have been automated if
kernel-wedge knew how to ignore a no more provided module (i guess the ?
symbol does this, right. Is there a reason not to make it automatic for each
module ? Do we expect to catch modules that dissapeared this way ? And if we
catch such, what should we do ?). 

The three other issues where issues that have undoubtly been fixed in some way
in the i386 package, and should really have been added to kernel-wedge, right ? 
These where the ones which involve deeper knowledge of how kernel-wedge works,
and will cause most trouble to porters which have only limited interaction
with d-i, and should i believe really be the job of the d-i team.

And Frans, no, insulting porters and complaining they don't do their job is no
way to get this solved, and i find joeyh remark that only 6 d-i architectures
support 2.6 kernels, while thanks to the common architecture, all debian
official architectures now have 2.6 kernels since a couple of months quite
illustrative of the problem faced by d-i here.

Sven Luther