transient vs committed types about h5netcdf

[shoyer]

@shoyer Linking the committed types seems to me to be the correct way of doing this. It has many advantages, eg. lesser filesize and easier discoverability. What makes me a bit nervous is why netcdf-c/netcdf4-python do not use this shared approach. I'm interested in your assessment of the above issue and like to hear your suggestion how to move forward.

This is a good question, I would suggest asking the netCDF-C developers.

[kmuehlbauer]

@ZedThree This might also be of interest for you.

[ZedThree]

My understanding is that netCDF files cannot contain HDF5 object references of any kind in order to avoid cyclic graphs. This means that definitions of datatypes (and dimensions as well) can only be searched strictly upwards in the group hierarchy (that is, types/dimensions for a variable in /group1/group2 will be first looked for in group2, then in group1, then in /, and nowhere else)

[kmuehlbauer]

@ZedThree That's my information too. The question is, why netcdf4-python is using transient types within the dataset instead of using shared committed type?

[kmuehlbauer]

I've opened discussion at netcdf-c Unidata/netcdf-c#2846.