The following snippet:

>>> u'@test-\u5171'.encode("mbcs", "strict")
'@test-?'

Should raise a UnicodeError.  The errors param is
completely ignored, and the function always works as
though errors='replace'.

Attaching a test case, and the start of a patch.  The
patch has a number of issues:
* I'm not sure what errors are considered 'mandatory'.
 I have handled 'strict', 'ignore' and 'replace' -
however, 'ignore' and 'replace' currently are exactly
the same (ie, replace)
* The Windows functions don't tell us exactly what
character failed in the conversion.  Thus, the
exception I raise implies the first character is the
one that failed.  For the same reason, I have made no
attempt to support error callbacks.

Comments/guidance appreciated.