Luc is exactly right about mmencode violating this. The bug was pointed
out to me about two months ago and is fixed in my private version, but
the fix won't make it to the rest of the world until the next metamail
release, which I'm still hoping will be next week. My fixes are almost
exactly what Luc prescribes.
The problem is indeed a subtle one. In my fixed version, I assume that
you want to map the local newline convention to CRLF for text/*, but not
for anything else, although this decision can be overridden for any
particular type via a mailcap entry (there's a new "textualnewlines"
option) and a new "-p" flag (for "portable newlines") in mmencode.
I apologize for getting this wrong in my implementation after we put so
much effort into getting it right in the spec. The surprising thing to
me, when it was pointed out to me, is how little trouble it actually
seems to have caused. I suspect this is because there is relatively
little line-oriented data being encoded in base64. (Indeed, the default
behavior for all my code is to use quoted-printable for text/* data, so
it probably happened pretty darned rarely.) -- Nathaniel