fedora(a)leemhuis.info (Thorsten Leemhuis) writes:
> > Problem: Users that want to compile kernel modules (Nvidia
Drivers...)
> > need to install the rpm.
>
> I do not see the problem here. Only few software can be built without
> additional packages (you need at least gcc, make, perhaps m4, bison,
> ...).
I see the problem. All those users of the nvidia driver will have
problems to build the kernel module.
Why? Do you think a user will see the difference between
| $ make
| bash: make: command not found
and a compilation error? It will be task of NVidia to write sufficient
documentation and/or diagnostic. And btw... the nvidia driver are not
supported by FC ;)
And what will they search for to solve the problem: a
kernel-source(code)
rpm that is will not be part of FC3 (as it seems).
This is caused by missing packing policies. When RH would require things
like
| Provides: %name-devel = %epoch:%version-%release
in packages with header-files but without a separate -devel subpackage,
a future splitting (e.g. because of efficiency reasons or additional
dependencies) would be much less painful. In this case, the
module-packages could buildrequire 'kernel-devel' and would build both
with monolitic and splitted packages.
> When the case above would be very common, you can make the
> header-package part of the base-system but without requiring it
> explicitly. So it can be disabled on minimal installations.
Don't understand this.
Add 'kernel-devel' to the comps.xml file (as 'type="default"').
Custom
kickstart installations could disable it by adding '-kernel-devel' to
the %files list.
How to solve the problem with updated kernels and the needed headers
for those?
??? Just add 'kernel-devel' to apt's 'Allow-Duplicated'; yum has
probably something similar. So it won't be a problem to install
multiple 'kernel-devel' packages.
Enrico