On Fri, 13 Aug 2004 09:56:58 +0100, David Woodhouse wrote: > I thought that conventional wisdom was that kernel packages should > always be _installed_ rather than upgraded, so that the old, working > kernel remains on the system in case of problems with the new one. > Especially as we make our kernels gratuitously fragile by omitting ext3 > support and _always_ requiring an initrd. > > Hence I was a little bit surprised when I upgraded my powerbook to > rawhide and the installer removed all the old kernels, leaving only the > latest kernel with a broken initrd that didn't boot. > > I filed a bug (#129640) and it was closed 'NOTABUG'. Apparently the > installer has always done this and always will. I still think it's a bug > though. > > If the general consensus is that the installer is correct, then we > should be consistent about it -- we should change up2date and yum to > upgrade rather than install kernel packages too. > > If the general consensus is that having an old, known-good kernel to > boot from in case of problems is a _good_ thing, then perhaps I should > re-open the bug. The installer ought to be correct and upgrade your installation to a working kernel. If it doesn't do that and the upgraded kernel/initrd doesn't work, it's a bug in the kernel or installer. On the other hand, if the installer kept your old default kernel and you would run first reboot with that one, it might cause problems and much grieve on your side. Default of upgrade installation should be to boot the upgraded system with the upgraded kernel, not a mixture of an upgraded system with some old unknown kernel. Btw, apt/yum/up2date are no official distribution upgrade paths. So what they to for normal kernel updates is something different.
