Note that pairing fsencode with 'utf-8' isn't guaranteed to do the right thing. It would work for the default C locale (since that's ASCII), but not in the general case.

Enhancing backslashreplace to also work on input is an interesting idea, but worth making it's own RFE: http://bugs.python.org/issue22286

I also agree we can ignore xmlcharrefreplace here.

So that leaves the basic pattern as:

data.encode('utf-8', 'surrogateescape').decode('utf-8', 'replace')
data.encode('utf-8', 'surrogateescape').decode('utf-8', 'ignore')
data.encode('utf-8', 'surrogateescape').decode('utf-8', 'backslashreplace')

This wouldn't allow the option of substituting an ASCII question mark, but I'd be OK with that.

Possible function name and implementation:

    def convert_surrogateescape(data, errors='replace'):
        return data.encode('utf-8', 'surrogateescape').decode('utf-8', errors)
    
Added bonus: pass "errors='strict'" and you'll get an exception if there were any surrogate escaped values in the string. (I take that emergent property as a sign that we're converging on a sensible design here)

Adding a fast path for keeping track of whether or not a string contains escaped surrogates would then be a separate RFE.