__FLOAT_BITS macro uses compound literal
i think this is the only case where
standard c header is not c89 compatible