is passed straight through to the unicode() and ustr.encode() calls.
I think it's the best we can do to address the UnicodeErrors in badly
encoded headers such as is described in SF bug #648119.