Steve Langasek wrote:
Um, I believe both instruction emulations are equally tested -- they're in the same patch.On Fri, Oct 01, 2004 at 01:07:25PM -0400, Nathanael Nerode wrote:So I produced a kernel intended for the upgrade-i386 directory.See http://lists.debian.org/debian-kernel/2004/08/msg02087.htmlI wanted it to be tested, but nobody has paid any attention. I don't currently have a place to upload it.How is this possible? The last discussions I had with the kernel maintainers showed that there was no patch available yet that securely implemented the 486 instruction in emulation that's needed by the new C++ ABI (as opposed to the known, tested instruction emulation needed by glibc).
This is subject to precisely the same security concerns as the packages currently in sarge and sid. After reading the relevant bug report, I am not quite sure what those security concerns actually *are*, however.
Well, upgrading the kernel patch is certainly valuable. I thought that working on the transitional kernels in parallel would also be valuable; any updated patch -- which will have to be present in the mainline kernels in sarge and sid -- can simply be plugged into the transitional kernels.I send this message to debian-release in hopes that someone will pay some attention, since the debian-kernel list apparently doesn't care about the upgrade-i386 problem.I know several of the subscribers care, but what's needed now is a sound kernel patch, not kernel packages dropped from above.
How, then, to address the problem that the upgrade-i386 kernel *must not* depend on initrd-tools, modutils, etc. from sarge? (If you can figure out how to address that in a package built in sarge, do tell.)Also, FWIW, the kernels you provided were based on the existing sources in woody. I believe (and the kernel folks I've talked to seem to agree) that it would be better if the upgrade-i386 directory could be based on kernel sources matching what we ship in sarge, since the two dists will have to be maintained in parallel for sarge's lifetime.
I can fiddle with the sources, of course, but as far as I can tell, it will still have to be built in a woody chroot, so what is the benefit?
In any case, having a minimally functional upgrade path can't be a bad thing; it's easier to make improvement when you have something functioning to test against.
Unless, of course, Debian decides to drop real i386s entirely -- which certainly has some arguments in its favor, and would obviate a lot of work. However, I believe that was decided against last time it came up.