Ben>  what do we do about translating the stuff into the
Ben> internal format when we read it in?
We don't worry about it. As far as no-mule XEmacs is
concerned, it's binary. Anything that is not IS0 8859/1
(sorry, world, I am not willing to go as far as /2 or /15) is
simply unrecognized, and should be protected by (featurep
'mule). Note that this doesn't mean that you can't treat a
buffer as ISO 8859/2 (or get "free" Euro support) by
declaring your font to be an iso8859-2 (resp. iso8859-15)
font, just that XEmacs doesn't translate Unicodes to ISO 8859 codes.
What I'm concerned about is Mule, not non-Mule. If we have UTF-8 elisp
files, then what do we do when we read in a UTF-8 string in a .elc file
under Mule? Which charset do we choose?
If we switch to UTF-8 internally, this problem goes away.