Allow user to select conda env for the dashboard

[danlester]

Multiple conda envs work correctly in Voila since the conda env name is embedded in the ipynb.

For the other frameworks, the user should be able to select from a list of available envs when they create the dashboard.

To try, on TLJH:

source /opt/tljh/user/bin/activate 
conda create --name newenv python=3.7 
conda activate newenv
conda install -c conda-forge ipykernel
python -m ipykernel install --name "newenv" --display-name "Python [newenv]"
conda install -c conda-forge dash # To show the env is different

If jhsingle-native-proxy adds something like this before the 'real' command:

conda init bash && conda activate newenv && ...

Then there is an error because the conda command 'eats' all of the command line and doesn't like e.g. '-m' of python -m.

simpervisor/process.py is usingasyncio.create_subprocess_exec instead of asyncio.create_subprocess_shell which could be the problem.

So this still needs some investigation.

[fcollonval]

I don't know if this can help. But nb_conda_kernels is using a special run script to execute the kernel within a conda environment.

More information:

• Kernel command: https://github.com/Anaconda-Platform/nb_conda_kernels/blob/master/nb_conda_kernels/manager.py#L195:

  spec['argv'] = ['python', '-m', 'nb_conda_kernels.runner'] + [conda_prefix, env_path] + spec['argv']

• Runner scripts: https://github.com/Anaconda-Platform/nb_conda_kernels/blob/master/nb_conda_kernels/runner.py

[danlester]

@fcollonval Thank you so much for the pointer! I was aware of nb_conda_kernels and felt the answer must be in there somewhere... Thanks for locating it, I hope - I'll try it out anyway.

Dan

[danlester]

This is now available in version 0.3.0 - thanks for the input @fcollonval !