Projects with both python and julia notebooks require some care with environments.
Environments and libraries installed in .julia/conda by Conda and accessed by PyCall,
which are in turn inside julia environments may not be compatible with
those in python environments.
Note python environments modify the paths in the shell, while julia environments are
specified inside the julia executable. It is possible but inadvisable to run julia from
inside python environments, as these may conflict with the recommended python managed
internally by Conda in .julia/conda.
~/opt/anaconda3/bin | the first vanilla anaconda jupyter
~/.julia/conda/3/x86_64/bin/jupyter | the jupyter for running julia notebooks, esp if they have to use Conda python libraries
~/mambaforge/envs/ASTRAL_soundings/{bin,etc,shared}/jupyter | for running python, each project should run in its own mamba/conda environment from the shell
In ASTRAL_soundings I use python packages from ECMWF et al. to read
BUFR files. My (shell-level) conda stopped being able to solve the environment, so I switched to mamba.
Mamba is basically the same as conda, only package environment are prechecked and so maybe install faster.
The main thing I needed was to reinstall a new package manager that didn't have a complicated base environment, then install
needed packages needed for each project in it's own environment.
Something in these environments didn't get along with .julia/conda. I commented out the mamba/conda initialization
that turns on the base environment in ~/.bash_profile and put it in ~/.local/bin/mambainit.bsh and ~/.local/bin/condainit.bsh.
Another suggestion is to use mamba deactivate to get back even out of the base environment.