Unicode format tests weren't actually testing unicode. This was probably due to the original backport from py3k.