On Mon, Dec 09, 2002 at 09:24:42PM +0100, Bill Allombert wrote:
> I do not agree. Installing a new kernel may require rebooting the machine,
> else you may have modules mismatch.
> Now, do you want that to happen noninteractively?
> In the pbuilder context, the modules installed are not the one used by the
> kernel, so it is not a problem. But this is specific to chroot environment
> which are not officially supported configuration, or else we need a mean
> to tell packages they are installed in a chroot. In fact this is exactly
> what do option 3) (or Herbert proposition) : tell the package that it is
> not in a standard setup. So either change pbuilder or design a standard way
> to query the 'chroot' status of a Debian install and send patches to the
> relevant maintainers.
> I think the solution is rather to not Build-Depend on a kernel package,
> because you should be allowed to build package without chroot and without
> installing a new kernel. Now I do not know how to do that in this case.
It would seem that we cannot resolve this issue here. (If a new
version of the kernel has already fixed this, I apologize).
kernel-image-2.4.19-i386-udeb-2.4.19 *needs* to depend on
kernel-image-2.4.19-i386 because it contains the modules already built.
Building them again seems stupid, wasteful of CPU resources, and prone
to error (eg. building with slightly different configuration).
So there are three options:
- pbuilder treats kernel-image as a special case, and sets things up so
it will install the kernel non-interactively.
- kernel-image respects non-interactive installs. Personally, I don't
see this as being that bad, if important feedback is required, then it
can be E-Mailed. If this isn't good enough for the requirements, then
chances are non-interactive approach should not be used anyway.
- kernel-image treats building under pbuilder as a special case, not an
approach I particularly like.
Brian May <email@example.com>