[CF-metadata] CF provisional standards

From: Egil Støren <egil.storen>
Date: Mon, 20 Nov 2006 16:04:32 +0100

John Caron wrote:
> Let me just appreciate that any individual interested in CF is already
> doing much more than is the "normal practice" of generating files with
> no published documentation at all. So even if some small group of data
> publishers creates some special thing, at least they documented it, and
> if its even on CF web page where I might know to look for it, they
> deserve thanks. So all the variant positions being argued here, IMO, are
> a huge improvement over usual practice.
>
> We have the "who pays vs who benefits" social dilemna here. Data writers
> must pay the price of doing the conventions right for the benefit of
> data readers. Data readers must pay the price of suboptimal conventions
> for the benefit of writers who need fast solutions, and dont want to
> rewrite their data. Depending on which role you play in your day job,
> you will have some visceral feelings on these issues. We need both POVs
> and the solutions will be tradeoffs.

Maybe it is more clearifying to consider three groups with different
interests:

1. Data writers
2. Writers of general software
3. Users of both data and software

What you and others mean by "data readers" seem to correspond to group 2
above. Naturally, this group has a noble goal of providing software that
can easily be used directly against files produced by the data writers.
This is a goal that is very difficult to obtain, and in general, users
have to close the gap between the actual files they must work with, and
the software tools available to them. This is done by scripting
languages and tools like NCO etc. I think most users are happy with this
situation, since it gives them a sense of freedom.

Accordingly, most uses of CF compliant files will involve a stage where
the user investigates the actual kind of files she/he will work with.
Although these files are CF-compliant, they may also obey some local
standards the user have to consider. Typically, the user will set up a
runtime environment for a) converting source files to some type of
destination files, and b) using general software tools for producing
results from the destination files. Considering this, the most important
thing about CF is that it is understandable from a human point of view.
A CF that is easy to implement in general software tools is a second
priority. We already have general software tools that work on the netCDF
level that will satisfy the needs of the users.

As a conclusion, I tend to look at CF primarily as a tool for data
writers to create files rich in metadata that are well understood by humans.

Best regards,

    Egil St?ren
    met.no
Received on Mon Nov 20 2006 - 08:04:32 GMT