[mpas-developers] Run Time I/O layer

[Doug Jacobsen]

Hi Michael,

Thanks for the comments. I also had convinced myself that it is completely
possible to have run-time create-able streams. The only reason I left it
out, is because I couldn't come up with a clean way to define the streams
at run-time. I did come up with a way to have fields arbitrarily added to
streams at run-time, but the only thing I came up with regarding streams
being defined at run-time was the example I put in the design document, and
that isn't really a "fool-proof" system of stream definitions.

I don't necessarily think the requirement that they be build-time defined
is a permanent requirement, and it could potentially be laxed in the
future. Also, I think if we have a good idea for creating new streams at
run-time it would be perfectly acceptable to have them defined at run-time.

So, in general I agree that it is completely possible to have run-time
definable streams, but I would prefer to have a clean setup for them before
I modified that requirement in the current design document. Maybe we can
brain storm about it sometime and see if we can come up with and clean
solutions. Also, I agree that it would make certain things (masking of
fields) easier as well.

Thanks,
Doug


On Wed, Mar 27, 2013 at 5:49 PM, Michael Duda <duda at ucar.edu> wrote:

> Hi, Doug.
>
> If we were to develop a new stream handling module, which I think makes
> sense, I'm able to convince myself that it shouldn't be much more
> difficult to allow for the definition of new streams at run-time. The
> mpas_io_streams module allows new streams to be arbitrarily
> instantiated, so presumably appropriate data structures could exist in
> the stream handling module to allow new streams to be instantiated based
> on the contents of a file read at run-time. Furthermore, the field data
> types contain the name of the field as a string, so there should be some
> possibility of scanning through all fields defined in the Registry until
> we find the one whose name matches that of the field to be added to a
> new stream object based on the contents of a file.
>
> So, I'd propose that we re-consider the requirement for build-time
> definable streams, and require that everything be run-time definable.
> This would make it easier, too, to define new streams in more
> sophisticated ways, e.g., by saying that a new stream 'foo' is a
> duplicate of an existing stream 'blah', but with a higher output
> interval and without some of the 3d fields in 'blah', or only over a
> high-resolution region of the mesh, which would have to be defined in
> terms of a mask field -- something that I think we're considering
> anyway.
>
> Michael
>
>
> On Fri, Mar 22, 2013 at 03:33:32PM -0600, Doug Jacobsen wrote:
> > Hello Everyone,
> >
> > As the XML Registry branch is now merged with the trunk, I want to
> continue
> > working on the run time I/O layer.
> >
> > Attached (and in the repository) is a copy of the design document.
> >
> > Any comments/questions are welcome.
> >
> > Thanks,
> > Doug
>
>
> > _______________________________________________
> > mpas-developers mailing list
> > mpas-developers at mailman.ucar.edu
> > http://mailman.ucar.edu/mailman/listinfo/mpas-developers
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/mpas-developers/attachments/20130327/1f2dec02/attachment.html