[Kst] NetCDF packages
nicolas.brisset at free.fr
Wed Oct 15 20:22:36 UTC 2014
Thanks to all for your answers.
To try and summarize all the information, here is how I currently see things:
- The original netCDF C++ interface we currently use sort of works, but with limitations which make it a bad choice in the long run
- The officially published rewritten netCDF-c++4 one is better, even if some further improvements are not yet publicly available (though I guess they will be at some point)
- If we make the switch, packages will follow for openSUSE and hopefully other distros thanks to the OBS and the PPAs maintained by some Kst users. For Windows we've been preparing it with Peter and I'm confident that we can pull it through, it's only missing cmake files for netcdf-c++4 but it's a small library so it looks feasible. And when we have that, building for Mac OSX should also be fairly easy
Under those circumstances, I think we should make the switch. I'm willing to adapt the netcdf data source to make use of the new library. Then it'll "only" be a matter of providing updated packages on all platforms, but I see no blocking point. Regarding the build system, as far as I could see we need to check for the following headers: netcdf.h for the C library and netcdf instead of netcdf_cpp.h for the C++ wrapper. Then we have to build a netcdf-c++4 lib and make some minor adaptations to the source code.
How does that sound?
----- Mail original -----
> De: "Lynton Appel" <Lynton.Appel at ccfe.ac.uk>
> À: "Nicolas Brisset" <nicolas.brisset at free.fr>, "Christian Trippe" <ctrippe at opensuse.org>, "Jan Sundermeyer"
> <sunderme at gmx.de>
> Cc: "kst" <kst at kde.org>
> Envoyé: Mercredi 15 Octobre 2014 11:11:15
> Objet: Re: NetCDF packages
> Hi Nicolas
> I developed a C++ API that I use in a code here at CCFE (Culham
> for Fusion Energy, UK) called EFIT++.
> This API is in the UNIDATA respository as you have noted. I had a
> problem working with user-defined types
> so I made some developments (improvements) on this which I never put
> back into the UNIDATA repository.
> At the time there was a bug in the HDF5 layer which took some time to
> resolve by which I time I had less time
> to work on the netCDF C++ aspect.
> So in summary my netCDF C++ API as published on the website does
> but I have a better local version
> that wasn't put back; also to my knowledge the original "official"
> that is on the netCDF website doesn't work.
> Hope that helps.....
> On 14/10/2014 20:55, Nicolas Brisset wrote:
> > Hi Christian/Jan,
> > You may have seen on the Kst list that we're trying to get NetCDF-4
> > to work better, which requires to compile NetCDF it with HDF5
> > support.
> > Under Linux, at least on OpenSuse, apparently there is HDF5
> > support. However, I tried to work with one file making use of
> > newer NetCDF-4 features and I noticed that the C++ wrapper we use
> > is obsolete.
> > Apparently, since version 4.2 the C++ library has been improved by
> > Lynton Appel (which I'm putting in CC in case he has news from the
> > packaging front) and split out of the base package, see
> > http://www.unidata.ucar.edu/downloads/netcdf/netcdf-cxx/index.jsp
> > and http://www.unidata.ucar.edu/software/netcdf/docs/cxx4/
> > Questions:
> > 1) do you know who's maintaining the NetCDF packages?
> > 2) do you think it would be an issue to update to this new wrapper?
> > 3) if it's feasible, how can we best phase the changes so that we
> > don't end up with a version with reduced support for NetCDF-4?
> > I hope the questions are clear, otherwise don't hesitate to ask for
> > clarifications!
> > Thanks for your help,
> > Nicolas
> This email is free from viruses and malware because avast! Antivirus
> protection is active.
More information about the Kst