1. Non-ASCII bytes were accepted after shift sequence.
2. A low surrogate could be emitted in case of error in high surrogate.