Separate coconut kernel installation and `nb_conda_kernel` support

Using nb_conda_kernels as my primary workflow (i.e. not needing a separate jupyter notebook server running for every kernel, just installing a kernel in each environment with one central environment running the server) I would

1. install jupyter, nb_conda_kernels, etc in the main/base/default environment, and run jupyter.
2. install ipykernel, project dependencies, etc in a project-specific environment.
3. open jupyter interface, and switch kernel. Not to “python 3”, etc, but to Python [conda env:projectenvname] which gives me all of my packages/dependencies needed for that project, still on the one jupyter server.

My question revolves around using coconut within this workflow (which is becoming more standard in teaching/academic circles afaik). I have coconut[all] installed in both the main and the “cocotest” project environment, and have run coconut --jupyter successfully in both as well. But I still cannot access coconut syntax directly (i.e. no magic functions) within a notebook using the Python [conda env:cocotest] kernel. Given that nb_conda_kernels explicitly supports other languages (ipykernel, r-irkernel, anything on that list of languages above) as long as the kernel is installed inside that environment, i don’t really understand why coconut isn’t working?

I did also try running coconut --jupyter notebook from the main environment, and selecting the Python [conda env:cocotest] kernel, but still no dice.

I think one solution is to either make a separate installable or maybe optional coconut[kernel] installation that will act like ipykernel, so that users can add coconut support in jupyter to their versioned dependency list (e.g. environment.yml, etc), just like they can now with ipykernel and r-irkernel, while only needing to run a single jupyter server at any given time.

[evhub]

@tbsexton Try again with pip install ---upgrade coconut-develop>=1.4.3-post_dev25 and I suspect it’ll work this time (also you shouldn’t even have to run coconut --jupyter now—it should automatically install the kernel in the right place for nb_conda_kernels on pip install).

I also just added a pyproject.toml that should give you the right setuptools by default—let me know if that’s working for you.

[fabianhjr]

Installing via pip install conda-develop[all] after removing the old coconut installation leads to an error unless i manually update setuptools in every conda environment (even if I start with a fresh one)

RuntimeError: bdist_wheel not supported for setuptools versions < 18 (run 'pip install --upgrade setuptools' to fix)

Can this version of setuptools be added as a dependency?

This might have been related to #560