On Wed, Dec 8, 2010 at 7:45 PM, Ken Moffat <[email protected]> wrote: > On Wed, Dec 08, 2010 at 06:31:31PM -0600, robert wrote: >> >> >> No, I got the "me hopes" part ... quite Shakespearean, in fact ... as in >> methinks ... >> >> It's the "build itself" part ... still don't understand what that means. >> Do you mean just write up a script and cut it loose to build the os? >> > The idea (in the days of "iterative comparison analysis", and > before), was that a new LFS system should have everything it needed > to build itself, and that the result ought to be identical (in > practice, after stripping files, and removing the different > compressed timestamps from gzipped files, and some other "we don't > quite know why this always differs" files). > > So, some of us used to take a new development or pre-release > version, and let it build itself (or rebuild itself in-place for > Greg's version of ICA). With modern toolchains, I've given up doing > that because there were too many unexplainable differences, perhaps > caused by address randomization. > > ĸen > -- > das eine Mal als Tragödie, das andere Mal als Farce > -- > http://linuxfromscratch.org/mailman/listinfo/lfs-support > FAQ: http://www.linuxfromscratch.org/lfs/faq.html > Unsubscribe: See the above information page >
The last sentance there hits the head on why ICA doesn't really work anymore. There's features in the kernel, glibc, binutils and gcc all of which will cause the build results to be slightly different each time. Heck, at work we're trying to figure out why our builds built over night run 2.3% faster than our builds during the day. And I'm not talking about the build itself, but the actual runtime. -- http://linuxfromscratch.org/mailman/listinfo/lfs-support FAQ: http://www.linuxfromscratch.org/lfs/faq.html Unsubscribe: See the above information page
