Now, I'm _not_ saying that the glibc change is wrong. In fact, it
enables extra gcc optimizations, which is great. But in this case it
looks like we're going to have to review all use of thread mutexes in
the whole of Fedora. Maybe not the kind of thing we had in mind for
Fedora 16 at this point.
I think it's great that Thomas Rast, Jim Meyering, and Jakub Jelinek
found the problem after probably a couple of man-days of effort, but
really development and bug fixing like this belongs in Rawhide.