Re the argument about numbers vs two-letter codes, vs names...
-- The name by which a language is called is politically and socially
sensitive almost anywhere. Whatever we do should, to the degree possible,
reference something externally developed and generally agreed on, not
try to invent our own rules. Whatever else already exists, "they" have
already shed the requisite liters of blood--there is no value whatsoever
in our repeating the experience. And, when people complain about how the
name of their language is spelled, why some other language got a "lower"
numerical code, or whatever, "blame it on ISO" is lots more attractive than
trying to figure out how to make them happy (or convince them that you
weren't discriminating against them out of some deep racism).
-- People who are concerned about what two-letter or numeric codes look
like to users are reminded that mail is sent from and received by User
Agents, not users. Display anything you like, use canonical and
unambiguous stuff on the network.
-- In this particular case, the "old UA" issue really doesn't apply.
While language tagging should be valid for, e.g., 8859-1 and, in the
parallel translations case, would be pretty useful, a lot of what we are
talking about here is 10646. If either is prepared by a MIME UA and
then sent over a seven-bit channel to an unmodified 821/822 UA, the
user is going to see garble and trash (much worse with 10646). Whether
language codes have to be looked up is really going to be insignificant
in comparison to figuring out what the message is trying to say. Of
course, there are exceptions to this, but making things less precise
in order to facilitate
text/plain; charset=us-ascii; language=pig-Latin
Doesn't impress me as wonderful engineering.
john