Re: Dependencies on kernel-image-x.y [was: NPTL support in kernel 2.4 series]
Martin Kittel wrote:
> B) most people compile their own kernels and don't bother registering
> those with dpkg, e.g. via kernel-package, and therefore their systems,
> -while actually running a suitable kernel- do not provide the required
> virtual package.
> With A) I absolutely agree, but I think B) is not a valid objection to
> having a dependency on kernel-image-x.y, especially since it is very
> easy to create a custom kernel package with kernel-package. Also the
> reasoning of B) could be applied to any package, starting with
> java2-runtime and then going all the way to libc.
The kernel is handled specially for a reason. You can e.g. test a
new glibc in a chroot, but this won't work for the kernel.
> But even with A) being a valid objection, I still consider it to be
> cleaner to state a dependency on kernel-image-x.y because
> 1) it explicitely and visibly states the dependency that is inherent in
> the package
> 2) the information the dependency provides is visible _before_ the
> package is even downloaded
> 3) on systems that only have kernel images of kernel version x.y
> installed and where those kernel images are properly registered (and I
> would argue that this is the majority of systems out there), having the
> dependency just works.
This means e.g. a kernel developer would have to install some otherwise
completely useless debian kernel in order to check if his MaxDB testcase
is still ok for the actually running version.