Just to be brief, but to express a vote.
I see Mark's message (with the encodings table) as exactly right. I
think I can simplify his final question, though (how to select an
encoding method for 8BIT text).
If an 8->7 translator really wants to be clever with 8BIT text that
might best be encoded either by QUOTED-PRINTABLE or BASE64, it can
simply choose the encoding that produces the shorter result--or
selecting based on any other favored heuristic. It might look only at
the first several lines or bytes, or it might simply look to see the
proportion of characters in which the 8th bit is on, or anything else.
It might simply always choose one encoding.
Why does the translator really need subtype knowledge? It can do a
``better'' job in some cases if it has that knowledge, but I doubt that
it's worth the trouble of specifying charset classes for all gateways
for all time. One can make good guesses knowing not the subtype of the
character set, but instead knowing the mechanics of the encoding
possibilities. Since the translator will have to know a lot about the
encodings it can make (like it will have to know how to make them), its
choice can be based on that knowledge rather than on some
harder-to-obtain knowledge about the charset encoding described by the
message source.
Craig