There was a panic where lexer returned None on malformed tokens. But now we just ignore tokenization errors in mbe.