Anaconda ignores version-specific kernel dependencies

[Ed Swierk]

As I mentioned a while back, I'm using a Kickstart script to point
Anaconda at a number of additional yum repositories, including
fedora-updates as well as a local repository of homegrown packages. I
then install just one package (Arora), which requires a bunch of
others for our development environment.

In particular, Arora depends on a specific kernel version (for
compatibility with a number of homegrown kernel-module packages):

  Requires: kernel = 2.6.18-1.2869.fc6

This worked fine as long as 2.6.18-1.2869.fc6 was the latest FC6
kernel. But since then, a newer kernel, 2.6.19-1.2895.fc6, has
appeared in the fedora-updates repository. Anaconda now really wants
to install the latest kernel package, ignoring the Arora package's
dependency on the older one. The result is a bit of a mess:

- only kernel-2.6.19-1.2895.fc6 is actually installed
- grub.conf contains entries for both kernels
- kernel-2.6.18-1.2869.fc6 is the default entry in grub.conf
(resulting in a "file not found" on boot)
- there is no initrd for kernel-2.6.18-1.2869.fc6
- the grub.conf entry for kernel-2.6.19-1.2895.fc6 has "-up" appended
(not sure if this means anything, as it ends up running in smp mode)

As a workaround, if I add an explicit %packages entry in my Kickstart
script, all is well:

  kernel-2.6.18-1.2869.fc6

(Actually, Anaconda still ends up installing
kernel-headers-2.6.19-1.2895.fc6, but I can live with that.)

So the point of this long-winded explanation is to ask whether what
I'm trying to do ought to work, or whether there is another way to
accomplish the same result: installing a particular kernel version
without having to edit my Kickstart script every time package
dependencies are modified.

It looks like selectBestKernel() in yuminstall.py is a tricky piece of
code, so I'd like to know whether this is the right approach before
fiddling with the logic.

--Ed