On Sun, 24 Sep 2006, Les Mikesell wrote:
On Sun, 2006-09-24 at 22:59, Matthew Saltzman wrote:
Yum shouldn't have removed the running kernel, so recovery
should have been a matter of hitting a key during reboot,
selecting the old kernel with the down-arrow key, and
hitting enter. Still, it would be nice if it didn't
break the system in the first place when it should be
moderately easy to detect that a new kernel isn't
going to work yet.
How would the kernel RPM know? It can't depend on the NVidia module.
How would anyone know? If lsmod says a module is loaded and
you can't get that module for a new kernel it's probably
not a real good idea to make that new kernal the default
for the next boot.
Well, *you* know because *you* installed the nvidia module package. But
modules come and go (and drivers can be modules or not), so no guarantee
that a module that you need sometimes is loaded when you do the update.
You can set UPDATEDEFAULT in /etc/sysconfig/kernel---that gives you some
manual control.
You could, if you knew yourself that a needed module wasn't
available. But you probably don't know that without using
something like yum to check...
Is that not what we're talking about? Yum updates the kernel. The kernel
RPM's scripts update /etc/grub.conf. The kernel RPM wouldn't install
without the nvidia module *if* it could include the dependency on the
nvidia module, but it can't. So
- you run yum
- yum updates the kernel
- you see that yum didn't get the corresponding nvidia module
- you take control (one way or the other) and don't boot the new kernel
- next time you run yum, you get the nvidia module
- so you boot the new kernel (and make it the default)
It might be useful for yum to allow the user to set up dependencies
manually. It might also be asking for a nightmare...
--
Matthew Saltzman
Clemson University Math Sciences
mjs AT clemson DOT edu
http://www.math.clemson.edu/~mjs
