Thanks for the explanation -- looks like I was way off base on that one.

I took a look at the code you provided but it doesn't work as a drop-in
replacement for _escape_cdata, since that function returns a string
rather than bytes.

However taking your code, calling it _encode_cdata and then refactoring
all calls _encode(_escape_cdata(x), encoding) to _encode_cdata(x,
encoding) seems to do the trick and passes the tests.

Specific example:

- file.write(_encode(_escape_cdata(node.text), encoding))
+ file.write(_encode_cdata(node.text, encoding))

One minor modification is to return the string as is if encoding=None,
just like _encode:

def _encode_cdata(text, encoding):
    # escape character data
    try:
        text = text.replace("&", "&")
        text = text.replace("<", "&lt;")
        text = text.replace(">", "&gt;")
        if encoding:
            return text.encode(encoding, "xmlcharrefreplace")
        else:
            return text
    except (TypeError, AttributeError):
        _raise_serialization_error(text)