Containers

[jon.bosak@sun.com]

NDR SC members,

I am not a voting member of the NDR SC, but as a concerned
observer, I'd like to urge that you consider withdrawing the
current rule regarding list containers at your meeting on
Wednesday.

Let me be clear that I share the feeling that we need more
containerization rather than less.  In fact, as I indicated during
the joint LC/NDR calls last week, the current example showing
multiple line items rattling around loose inside the invoice
document makes me very nervous.  I believe on general principles
(*very* general principles) that this is going to cause problems
for processing applications farther down the line, and the
excellent position paper developed by LC on this subject has not
convinced me otherwise.

That paper has, however, convinced me (1) that the proposed cures
are worse than the disease and (2) that we don't understand the
potential problems well enough to propose a better solution at
this time.  Furthermore, it appears that there are no currently
demonstrable performance consequences large enough to offset the
additional complication that would result from the generation of
list containers.

With regard to the proposal (which I believe was mine) to assign
LC the task of manually indicating list containers in the data
model, the reply that LC has already identified the semantically
meaningful containers seems to me a valid one.  LC is saying, in
effect, that they have already containerized everything that made
sense to put into a container, and that if I'm bothered by the
sight of line items running around naked, then the onus is upon me
to provide a good semantic reason for categorizing the group as an
ABIE.  And I don't have one.

I've listened pretty hard to the recent discussions on this point,
and my conclusion is that we will have to revisit this question in
some future major (non-backward-compatible) revision, but that we
don't undertand the problem well enough right now to design a
clearly appropriate solution.  No one has questioned that the
schemas will work if they remain as they are now without the
application of the list container rule.  Since it appears obvious
that the application of an automatic container generation rule
will significantly complicate the schemas, and since we cannot
show a clear benefit commensurate with this additional complexity,
I think (as an interested observer who cares about our time line)
that we should proceed without nonsemantic containerization until
experience can show us a better way.

Jon