On Tue, 26 Nov 2002, Hamish Moffatt wrote: > On Tue, Nov 26, 2002 at 10:41:24AM +0200, Martin-Éric Racine wrote: > > On Tue, 26 Nov 2002, Ralf Hildebrandt wrote: > > > Inteesting question. One could install a kernel by hand (not as a > > > package). Thus a runtime test would probably be more suited. > > > > The problem with a runtime test i.e. in the package's pre-install script, > > is by > > the time you try it, the libc binaries have already been unpacked and have > > probably overwritten the older ones. > > Debian preinst scripts are run before the package is unpacked.
Correct. > However I looked in libc6's preinst and postinst and couldn't find this > test at all (2.3.1-5). Indeed. There should have been... :( > > 2) if there really is no other way e.g. because recent bug fixes in the > > pa-risc > > kernel branch were performed, then the libc6 package should at least depend > > on > > some specific kernel-image version (or newer), to prevent disasters. > > > > Besides, this is Debian; we have the kernel-package and people should use > > it. > > Therefore, if this new libc6 depends upon specific kernels, it should not > > be a > > problem to have the package actually mandate that, by having > > kernel-image-2.4.19 > > or newer as a dependency. > > There's no requirement to user a standard kernel-image package or even > kernel-package to build the kernel. However, the libc could Conflict > with kernel < 2.4.19 and that would help at least some percentage of the > users. That's my idea as well, to prevent disasters at least for those who _do_ use kernel-packages, their own or those by the Debian Project. :) -- Martin-Éric Racine "Kas sa tahad mind? - Nej!!! Är du en idiot?!!" http://www.pp.fishpool.fi/~q-funk/ Tallinn, Eesti; Espoo, Suomi.