Last week, on Wednesday, the Unidata netCDF team spent the day with Quincey and Larry of the HDF5 team. This was great because we usually don't get to spend this much time with Quincey, and we worked out a lot of issues relating to netCDF/HDF5 interoperability.
I came away with the following action items:
- switch to WEAK file close
- enable write access for HDF5 files without creation ordering
- deferred metadata read
- show multi-dimensional atts as 1D, like Java
- ignore reference types
- try to allow attributes on user defined types
- forget about stored property lists
- throw away extra links to groups and objects (like Java does)
- work with Kent/Elena on docs for NASA/GIP
- hdf4 netCDF v2 API writes as well as reads HDF4. How should this be handled?
- John suggests not using EOS libraries but just recoding that functionality.
- HDF5 team will release tool for those in big-endian wasteland. It will rewrite the file.
- should store software version in netcdf-4 file somewhere in hidden att.
- use HDF5 function to find file type, this supports user block
- read gip article
- update netCDF wikipedia page with format compatibility info
- data models document for GIP?
I have been assured that this blog is write-only, so I don't have to explain any of he above, because no one is reading this! ;-)
The tasks above, when complete, with together add up to a lot more interoperability between netCDF-4 and existing HDF5 data files, allowing netCDF tools to be used on HDF5 files.