Install non-Python dependencies with dnf

[yhtang]

Hi there,

This is related to the 'installation' checklist item of the JOSS review:
openjournals/joss-reviews#4149.

Could you please also provide the external package names under the DNF/RPM system? This could be very helpful for users of Linux distros such as Fedora, CentOS, and Scientific Linux, etc.

Many thanks and please let me know your thoughts!

[mortele]

Hi @yhtang, thank you for the suggestion. I have briefly looked into making the package installable (wholesale with all the dependencies packaged) with some of the standard package managers like anaconda and spack. This is on the todo-list but has been pushed back in favor of more pressing matters at the moment.

In all honesty, In my own natural Debian bias I had not even considered making a yum package, but I will investigate how to make it RedHat installable when I get some spare time soon.

[yhtang]

Oh, I guess there is some misunderstanding here. I am not asking for packaging the software as an rpm package. Rather, I was just wondering what would be the DNF counterparts for the following installation steps from READMD:

sudo apt-get update -y
sudo apt-get install -y pkg-config libhdf5-mpi-dev
python3 -m pip uninstall h5py  # Remove any serial h5py installation present
CC="mpicc" HDF5_MPI="ON" python3 -m pip install --no-binary=h5py h5py

Update: especially the first two lines, since the rest are most likely Python-centric.

[mortele]

@yhtang Aha, ok, my mistake! The following builds the non-Python dependencies in a fedora:35 docker container and so essentially answers the question:

dnf install -y openmpi-devel hdf5-openmpi-devel python3-devel python3.8
curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
python3.8 get-pip.py
export CC=/usr/lib64/openmpi/bin/mpicc
python3.8 -m pip install cython numpy mpi4py
export LD_LIBRARY_PATH=/usr/lib64/openmpi/lib/
HDF5_MPI="ON" python3.8 -m pip install --no-cache-dir  --no-binary=h5py h5py===3.6.0

I will have to work out tomorrow how to properly build h5py-mpi, which seems to be a small issue here, and add it to the docs as a separate installation guide section.

Seems like the default Python 3.10.2 in fedora:latest does not play well with h5py, so falling back to Python 3.8.12.

[yhtang]

Seems like the default Python 3.10.2 in fedora:latest does not play well with h5py, so falling back to Python 3.8.12.

I wonder if a virtual environment-based recipe could make life slightly easier? 😁

[mortele]

Hi again @yhtang, the documentation should now be updated with instructions for installing HyMD using apt, dnf, or homebrew, see #172 .

I did not add virtual environment-based installation steps, because I mostly expect users to know themselves when/if that is a good idea for their specific use case. I did however note that on Fedora, the default Python3.10 has some issues compiling MPI-enabled h5py, so maybe it would be good to add a small pyenv guide.