It is read in as a byte string on Python 2.x, and we don't do any decoding. We could just say its alwaysutf-8. Or, for bonus points, read a-*- encoding -*- marker and use that.