sh: 1: getfattr: not found when loading dataset in RTD

[zmoon]

I recently started seeing sh: 1: getfattr: not found in my notebooks on Read the Docs in the output of cells that load a netCDF from disk, for example, here (since fixed).

Another example in the wild

https://projectpythia.org/unstructured-grid-viz-cookbook/notebooks/recipes/mpas-atmo/#relative-humidity

ncdump -h does it too, before showing results. (this is a nompi libnetcdf)

The xarray docs use nbsphinx instead of myst-nb, where I guess it doesn't show up.

---

Now it does still work, but I assume this should still be reported somewhere, I'm just not sure where (RTD, xarray, netcdf4-python, netcdf-c, myst-nb, theme, ...).

[zmoon]

Looks like this might be the cause conda-forge/netcdf4-feedstock#174. I guess in the libnetcdf build environment getfattr is available¹ but then on RTD it's not. Some potential solutions:

• use your RTD config to get the attr APT package installed in your docs build environment, or install from conda-forge
• use h5netcdf instead

Footnotes

1. I'm a little confused since it doesn't seem to be there in the forge image. However mamba repoquery depends -t libnetcdf --channel conda-forge --remote suggests mpich is what leads to attr getting installed. But this is still happening with the nompi libnetcdf. Ah, but mamba repoquery depends 'libnetcdf=4.9.3=nompi*' -t --channel conda-forge --remote shows hdf5 is still bringing in mpich. ↩