On the other hand, if I have a MIME HTML message where the first part is
text/html with reference to part 2, for instance "here are the <a....>
financial reports</a> for this quarter", and then copy the first body
part to a new message for the next quarter, I want the reference to be
to *this* message's body part 2, not the previous quarter's.
Interesting counterexample. However, I'm not sure there is really any
sort of problem here. It seems to me that there are two cases:
(1) You are using a very stupid composition agent, in which case you are
likely to be simply pasting the content in without mucking with the headers
at all. The content-ids will not change (which is bad -- they should be
updated), but they will still match.
(2) You are using a smart composition agent. A smart composition agent will
pull in the message in its entirety and when you replace the second part
with a new one it will adjust the content-id and HTML reference fields
accordingly.
It seems to me that the only problem arises with a broken composition agent,
which handles only a portion of the replacement operation. This is clearly
wrong, and you don't have to be working with email or relative references for
it to be a problem.
Also consider that such a broken composition agent isn't likely to handle the
insertion of a new part between two existing parts correctly either.
Finally, note that in all cases we're talking about generating a completely new
message. The old message is just a source of raw material -- it isn't being
forwarded or anything like that. The handling of linked HTML source material
for purposes of generating new linked HTML material is a very complex problem
independent of the particular context we're talking about here.
Ned