Before CodecInfo tuple subclasses were introduced, the lookup() function
used to return a tuple: (encoder, decoder, stream_reader, stream_writer)

See http://www.python.org/dev/peps/pep-0100/

These were normally addressed by position rather than name and still are
in most cases.

I'm not sure whether the choice of attribute names on CodecInfo was done
on purpose or a simple oversight. Walter added them... I've added him on CC.

They do match the names of the methods on codecs, so perhaps we should
just change everything to read encode/decode instead of encoder/decoder ?!