Future of NeXus

[Ray Osborn]

on 2000/06/20 1:40 AM, Mark Koennecke at Mark.Koennecke at psi.ch wrote:

> 
> Ray,
> 
> hold on! Within a european SCANS project there will soon be a person to
> work halftime at NeXus. Hers or his first project should be a
> reimplementation of the NeXus-API in HDF5. At least that is what I
> thought. I would have placed a message with our plans into this mailing
> list as soon as we know when this person actually starts for everybody
> to object or aprove. When will soon be? We have a candidate, actually I
> believe even two, but the swiss authorities are very slow in confirming
> the position.   
> 

That's great news.  I didn't want to sound too aggressive about moving to
HDF5 because I thought you worried about jumping too soon.  However, if you
are going to get support for this work, then I shall definitely give SNS the
commitment they requested.

Please let me know the official position when you know it so that I can
update the web site.

I think the HDF5 port should be reasonably transparent to the external user
although we will have to rethink how we link data to more than one group.
At the moment, we store tag/ref pairs for subsequent linking whereas HDF5
uses absolute path names for unix-style linking.  I'm not sure if we can get
away with just redefining the NXlink structure as a character string.

> The issue of third-party tools won't be so serious
>> once our own tools become better developed.  For example, do we need IDL
>> support for HDF5, or can we implement Mark's IDL interface as external
>> procedures?  
> 
> If RSI is too slow with HDF5 support we can use IDL's C-API for NeXus5
> support. This can become maintaince nightmare though as we would need
> to maintain shared libraries for a couple of platforms.
> 

I guess the problems are no worse than maintaining the NeXus library itself,
but you are right that it means more work.  I'm more worried about all the
other programs as well (Data Explorer, Matlab, ENVI, HDFLook...).  At the
moment, it's very good publicity for NeXus that we can automatically read in
neutron data into these programs.  However, if NeXus becomes more widely
accepted in the neutron/x-ray community, that won't be such an issue.

>> On the other hand, there isn't even a VMS version of the HDF5
>> library yet, so I'm not saying we make the transition tomorrow (actually, we
>> might have to do that one ourselves - any volunteers).
> 
> Are there many people insisting on this? VMS is slowly being phased
> out.......
>

That may be true by now.  Both ISIS and IPNS still have a considerable
number of VMS machines, but I think both facilities increasingly use other
platforms for their data analysis.  Without an HDF5 library, they can't
store their data on VMS front-end computers, which could be a problem.  On
the other hand, it may not be difficult to link the HDF5 library on VMS -
it's just that HDF don't provide precompiled binaries yet (if ever).

Mind you, I read somewhere that Compaq is planning to market VMS more
aggressively again but it's probably too late for the scientific community.

>> I hope that those of you who attended thought that NeXus was
>> well-represented at NOBUGS.  I've read the abstracts, and would like to
>> update the web pages with some of the new stuff you presented e.g. on the
>> XML proposal, web servers at PSI, and the NIST open source effort.
> 
> I'll send you something. At NOBUGS it turned out that people feel a need
> for a Java-NeXus-API. We probably have to do that on top of NeXus and
> JNI. 
> 

The fact that there is not a native java implementation of HDF is a major
stumbling block here at IPNS, which has become addicted to 100% caffeinated
java.  However, they are still going to adapt their software to access the
NeXus format because they see it's going to widen their potential user base.
How much work would a java API be?  It would be the first object-oriented
implementation.  Would that require a rethinking of any of the design?

> Furthermore I think it is time to have another SoftNESS meeting in order
> to nail out a procedure for verifying instrument type descriptions or a
> more complete glossary. Perhaps we use Chris XML ideas for verification.
> 
> 

I'm a big fan of the XML idea.  We will definitely organize a workshop this
autumn.  I will ask SNS if they are prepared to provide any funding.

Regards,
Ray
-- 
Dr Ray Osborn                Tel: +1 (630) 252-9011
Materials Science Division   Fax: +1 (630) 252-7777
Argonne National Laboratory  E-mail: ROsborn at anl.gov
Argonne, IL 60439-4845