JakeVanAdrighem added a comment. I think it was able to avoid test failure simply because there are so few elements in the existing test that `compareNames` in qsort just happened to produce the same results each time. http://reviews.llvm.org/D12851