https://bugzilla.redhat.com/show_bug.cgi?id=1221824
--- Comment #27 from Filip Andres filip@andresovi.net --- My current hypotheses: * It's either a build problem or an upstream problem. * A build problem sounds a bit more probable to me as I would expect somebody else to hit the problem ahead of us. * If it's an upstream problem then I found the following:
As per ethr_atomics.c in the OTP source code there are two implementations of some functions -- one for architectures with native support, the other for architecture without it:
/* * This file maps native atomic implementations to ethread * API atomics. If no native atomic implementation * is available, a less efficient fallback is used instead. * The API consists of 32-bit size, word size (pointer size), * and double word size atomics. ...
Given the fact that packages built on my newish i7 processor work for me and packages built in koji don't, hypothesis is that there may be something fishy in the fallback branch of the code.
I've noticed that the build servers used by Koji use Atom processors (or at least the compiler for i686 builds is run with the -mtune=atom switch). Is it possible that the the processors lacks double word atomics, thus the fallback branch is run and there is an error there?
Is it possible to force a i686 build on a standard Intel processor not an Atom one?