Logged In: YES 
user_id=21627

Found it. PyUnicode_FromEncodedObject converts the string object to 
char*/len, then calls PyUnicode_Decode. This special-cases UTF-8, Latin
-1 and ASCII, then creates a buffer object and passes it to 
PyCodec_Decode.

Even if it might be possible to pass the string directly to the codec, the 
codec still has to deal with buffer objects, for direct callers of 
PyUnicode_Decode. So I leave the fix as-is, added a test-case 
(test_codecs.py 1.10), and close this as fixed.