Red Hat Bugzilla – Bug 914019
gcl: FTBFS in rawhide
Last modified: 2013-03-23 16:08:56 EDT
Your package gcl failed to build from source in current rawhide.
NOTE: build logs are cleaned up after 1 week
For details on mass rebuild see https://fedoraproject.org/wiki/Fedora_19_Mass_Rebuild
*** Bug 914432 has been marked as a duplicate of this bug. ***
The gcl package has consistently built successfully in mock for me, ever since the mass rebuild. It also consistently fails to build on the koji builders, complaining about memory corruption. I don't yet know what is different about the koji builders that could cause this.
No, that's not right. What is *really* going on is that the build succeeds on x86_64 (which is what I've been building in mock), but fails on i386. The same thing happened with gmp-ecm, and that turned out to be due to changes in GMP 5.1. Since GCL built successfully in mid-January, before GMP 5.1 was imported to Rawhide, it seems plausible that some adjustment needs to be made to the GMP-using code in GCL. I will try to track it down.
It turns out that adding -fno-omit-frame-pointer "fixes" the i386 build. I still don't know why that is, but it has nothing to do with GMP. Without that option, the value stack overflows, stomping all over any memory in its path. I'll do a build tonight with that option, and work with upstream to discover why it is necessary.