On Wed, Nov 14, 2007 at 06:44:20PM +0500, Alexander E. Patrakov wrote: > 2007/11/14, Bryan Kadzban <[EMAIL PROTECTED]>: > > My initial reaction is to use either current or 2.6.2, with 2.6.2 > > (slightly) preferred. > > -1 for "current", because this way two users, one starting from Fedora > and one from Ubuntu, will end up with different glibc binaries. > But does that matter ? For an individual's builds, identical biaries are good. But for two different builders, why would it matter when we don't expect anybody to distribute the compiled versions ?
Thanks for the reference to where to look, I can now see that I've been wasting my time building agaisnt test kernels with --enable-kernel=2.6.0 (e.g. no *at syscalls, private futexes, fallocate - all the recent changes which sounded like good things to have). So, for me +1 for "current". Maybe at this point I ought to mention that I've always been in favour of building the kernel I'm going to use in my LFS (and clfs when chrooting, the 'boot' variant there necessarily builds a kernel before the final system) and then using that to build the new system and run the tests. We need to give a bit more information to our builders so that they can make an informed decision. I think we have 4 categories of users for this discussion: 1. Existing users now building a newer LFS from their old one. They can easily build the new kernel first. 2. People using an LFS Live CD - the kernel version normally matches the sources on the CD. 3. Other users who can build a kernel suitable for their system, and should perhaps be encouraged to do this before building LFS (for one thing, it separates any issues in their .config from issues caused by attempting to follow the book). 4. People who cannot build a suitable kernel until LFS is completed (e.g. people following the book with the correct downloaded package versions, but booting from e.g. an ubuntu or knoppix Live CD, or people whose distro requires them to build a kernel with an initrd or initramfs. Note that some of these people may even be running a _newer_ kernel (e.g. rawhide, if that uses initramfs or whatever). I suggest that the command in the book should be '--enable-current' and therefore testers using jhalfs will use that. The first time it is mentioned, it should be followed by an explanation, with a warning that earlier kernels than the specified version cannot be used, and a further note for testers that building on an old kernel with --enable-current and then using the new system to build itself with the same flag will result in a different glibc, and if they need repeatability they should specify the old kernel version in the second set of builds. We should also warn people that they will be unable to use a kernel older than the version they specify. And a big warning for people with a _newer_ kernel than is in the book that they should use --enable-kernel=x.y.z (to match the book's kernel version) otherwise they will not be able to use the book's kernel. ĸen -- das eine Mal als Tragödie, das andere Mal als Farce -- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
