Would this patch already solve? :

https://github.com/python/cpython/pull/19130

There seems to be another bug: The strange 'latin-1' default encoding of cgi.parse(), which only has effect in non-mulitpart:

    if hasattr(fp,'encoding'):
        encoding = fp.encoding
    else:
        encoding = 'latin-1'


( cgi.FieldStorage and the other functions in cgi and urllib.parse use a 'utf-8' default correctly - and do not try fp.encoding, which is usually not present and not reasonable in form handling WSGI. And 'application/x-www-form-urlencoded' implies  utf-8. )

=> that default should possibly become utf-8. Optionally cgi.parse() could take an extra parameter encoding=None  .