[CF-metadata] proposed rules for changes to CF conventions

From: Jonathan Gregory <j.m.gregory>
Date: Thu, 28 Jun 2007 15:37:11 +0100

Dear all

I believe that in Paris there was agreement that we needed to be able to make
decisions on reasonable timescales. What has often happened is that a good
discussion has taken place, but with no agreed procedure to make a decision,
once things go quiet the issue drifts into limbo. We have several such issues
from the last several years. That was the reason for my making the proposal,
which I thought we had essentially agreed in Paris, but of course not everyone
was there.

Most requests for extension to the conventions come from people who intend to
write data fairly soon and find that CF lacks a convention for something they
need. Someone who makes such a proposal might indeed be the only person who
sees the need for it at that time, but that in itself is not a good reason to
reject the proposal. There is a first time for everything! What is necessary
is that other people pay some attention to the matter, and form a view. That
might not be a lot of people. We discussed in Paris that not everyone feels
able to participate in all discussions because of expertise. I proposed that
it was good enough if just the moderator and the proposer had thought about the
proposal, but I agree it would be safer if we required there to have been one
other person express a positive view during the discussion. Note that a single
person posting to express a reservation is sufficient for the moderator to
have to try to engineer a consensus or arrange a vote if this has not been
answered (because in that case there is an outstanding objection). That is
already quite a large safeguard, I would say.

Is that OK? I don't think it is necessary to have as many as five. Given past
experience I think few issues would be agreed with such a requirement. For
instance, on several previous occasions I have appealed for people to express a
view (positive or negative) on a particular matter under discussion and had no
response. To solve this kind of problem, greater engagement is needed, not
tougher rules, I'd say. It is necessary for people who think CF is valuable to
engage in discussions about it at the time changes are proposed.

I think the requirement for test data should be included in the requirement for
testing, rather than at the time of the proposal. Proposals always include
CDL examples. A real file may find problems once it is given to software as
input, and obviously it must exist for the testing to take place (so I agree
we need to require such a file), but it will not help the discussion of the
proposal.

The idea was the the libcf and the CF-checker are the two applications which
are required to verify that the new convention "works", in the sense that it
can be successfully interpreted. I think that's a pretty good indication it is
not flawed and does not break applications. To require two further applications
to use the data would, I think, be a recipe for no convention ever passing out
of provisional status on a foreseeable timescale, since those two are the only
ones which have a stated commitment to support CF.

It's a good suggestions that the committee could decide what to do if an
agreed change fails at the stage of testing. Alternatively, it could be
referred back to the proposer by default.

As I often do, I am arguing from the point of view of a data-provider. I see
the need to design conventions carefully, but if the standard cannot be
updated on a timescale which matches the needs of people who want to use it to
write data, it's not a lot of use. They will just invent their own conventions
and that will frustrate the purpose of CF.

Best wishes

Jonathan
Received on Thu Jun 28 2007 - 08:37:11 BST