Right now, I'm thinking that newline canonicalization should be done ONLY
under these circumstances:
content-type content-transfer-encoding
text/* all encodings
anything else quoted-printable and base64
In terms of what agents must do this is all quite correct. However, these cases
are not at all the same -- only the first constitues canonicalization.
Moreover, the first case is done "under" the encoding, in that the data is
change before the encodings are applied. In the second case it is the encoding
itself that is "canonicalized" into the local text form for encodings, not
in terms of data formats.
That is, when transmitting a content type of anything besides text, the ONLY
time you touch the newlines is when you're dealing with a transfer encoding
type of quoted-printable or base64, and then only while the part is encoded.
This should be specified more clearly in the MIME RFC's.
Well, the second operation is really beyond the purview of the specifications,
in that it gets into the question of how mail messages are canonicalized as
a whole for storage in the local environment. I'll see if I can clarify this
somewhat, but not at the expense of getting into message store format details.
Ned