Hi, Troy Benjegerdes writes:
> It may be efficient for maintainers, but it leads to non-x86 arches > being second-class citizens if everyone has to wait for the x86 > maintainer to prepare arch-specific kernels. The result is I just go > download source and build for all my PPC debian machines. I'd really > rather get a debian package. But if ppc has to wait for x86, it just > gets to take too long. Reality check, using 2.6.6 as an example. The upstream kernel was released to kernel.org on Monday 10th. Herbert's kernel-source package appeared on people.d.o on Wednesday 12th, well before the i386 kernel-image packages. The powerpc kernel-image packages were uploaded to ftp-master.d.o on Thursday 13th, and simultaneously made available in an inofficial repository. They finally appeared in unstable on Monday 17th. Now please tell me how this was too long, where precious time could have been saved, and what changes to the kernel packages in particular and to Debian in general this would require. > I'd like to propose we attempt to build x86, amd64, ppc, and ia64 > kernels from the same source tree. Your proposal has first- and second-class archs in it as well. And it means that several machines would be required in order to build all binary packages from that one source package, even if just a small arch-specific change was made. I won't even speak of a sufficiently large testbed, which for PowerPC alone involves half a dozen different machines at the moment. I'd like to propose we create a source package that only builds the binary package kernel-source and nothing else, and create the arch-specific kernel-image packages from kernel-patch-<arch> source packages. Of course, the patch would be empty if the arch was well-supported upstream, and then kernel-image-<arch> is probably a better name for the source package. In fact, this is how it's done now. Regards, Jens. -- J'qbpbe, le m'en fquz pe j'qbpbe! Le veux aimeb et mqubib panz je p�zqbpbe je djuz tqtaj!
