Hello Steve,
> I believe that we (CF, that is) have agreed upon a
> process that states that before a new feature can be added to CF, 1)
> reference files must be created and 2) clients that can read this
> data
> must be created and tested.
>
I think you are right - that was the kind of procedure that was outlined
in the white paper. I think this is a good idea in theory, I was really
enthusiastic when I saw it in the white paper - but now I'm not sure how
it works in practice. I think my main concern is how cf-comprehensive
the test application has to be before you can change something from
'provisional' to 'official'. Does a stand alone, does one job only,
application count, or do you need tests in larger application?
For instance there is currently an agreement (I think) to add the
'realization_weight' standard name qualifier to the CF standard.
(
http://www.cgd.ucar.edu/pipermail/cf-metadata/2006/001141.html). I
guess this is 'provisional' at the moment. I could probably create some
files based on a time series of a the global mean including realization
and realization_weight - and I could submit an IDL program to process
these to produce a 'probabilistic forecast' plume (in a fairly short
time frame). Would this be enough? Its testing the addition in a very
limited context? (looking back at the white paper I think it implies
more than this). So how do we agree what testing is needed for anyone
particular 'provisional' addition?
I'm also a bit uncomfortable about the implications of having to fix
files with rejected provisional content - especially if I have
distributed data to other users. I guess the way out of this would be
to publish 'fixes' (in most cases, hopefully, a set of ncatted's??) on
the CF-web site somewhere?
Jamie
Received on Wed Nov 15 2006 - 03:18:16 GMT
