pynio installation downgrades esmf/esmpy to no mpi versions

[adamsg69]

I have an environment where esmf and esmpy are installed with MPI support.
When I attempt to install pynio (conda install --channel conda-forge pynio), I get notified that many of the libraries I have already installed will be updated or downgraded to versions without MPI support:

libnetcdf 4.7.4-mpi_mpich_hfd9c5b6_5 --> 4.7.4-nompi_h9f9fd6a_101
netcdf-fortran 4.5.3-mpi_mpich_h3923e1a_0 --> 4.5.2-nompi_h45d7149_104
netcdf4 1.5.4-nompi_py38hec8b9af_102 --> 1.5.3-nompi_py38heb6102f_103
hdf5 1.10.6-mpi_mpich_ha7d0aea_1 --> 1.10.5-nompi_h3c11f04_1104
esmf 8.0.1-mpi_mpich_h213fab7_100 --> 8.0.0-nompi_hb0fcdcb_6
esmpy 8.0.1-mpi_mpich_py38h08b7006_100 --> 8.0.0-nompi_py38hf0e99fa_1

However, disabling MPI support disables some functionality I use that depends on MPI.
How can I install PYNIO without these packages loosing MPI support?

Thank you,
Adam

[naomi-henderson]

I am having the same issue - I need both mpi (currently required for the xesmf package) and pynio. I could split my workflow and use separate kernels, but it makes it much harder to share code and kernels with colleagues. Has anyone tried installing from source code to see if nompi is really required here?

[JimmyChen277]

I am facing the same problem. It really confuses me.

[jduckerOWP]

I am also facing this same problem here. Has there been any resolution at this time for installing PYNIO without esmpy and other packages loosing MPI support? This basically prevents users from opening meteorological forcing files (e.g. GFS grib2 files) and allowing us to regrid these forcings to other datasets using ESMF regrid capabilities. This is a huge upcoming need in the scientific community for python users. Any help here would be greatly appreciated. Thanks!