Hints to quarto of where to look for a jupyter executable?

[dmbates]

My colleagues and I are very much enjoying using Quarto for literate programming with Julia (and looking forward to this Juliacon presentation by @jjallaire on Friday).

If a Julia user doesn't have jupyter installed when they add the IJulia.jl package a Conda instance (currently the miniforge flavor IIRC) will get installed in their ~/.julia folder (Mac and Linux - not sure where it is under Windows) with the jupyter executable at ~/.julia/conda/3/bin/jupyter.

What would be a good way of having this location recognized as jupyter in the quarto configuration? For myself I just create a symbolic link from my ~/bin directory, which is on my PATH, to this executable and everything is cool. But I am writing instructions for students in a short course in September and they may not be the sort who are thrilled when asked to "create a symbolic link"

[tpoisot]

I second this entirely! If possible, it would make sense to use the jupyter version that is bundled with IJulia.jl, rather than having multiple installations on the same system.

[jjallaire]

You can define a QUARTO_PYTHON environment variable which will override our internal checking of PATH, etc. for the version of Python to use.

[dmbates]

@jjallaire I can check this, of course, but is there a similar environment variable for jupyter? My concern is with the evaluation of Julia code blocks after a .qmd file is converted to a Jupyter notebook. Is it assumed that the jupyter executable will be in the same directory as the python executable?

[jjallaire]

We actually don't invoke Jupyter via the executable (which is no longer recommended due to PATH pollution created by the many versions of Python in play on a system). Instead we do the following when discovering kernels:

python3 -m jupyter --paths --json

That's the only time we actually invoke the jupyter module/executable. All of the rest of our interaction is via the Python packages nbformat, nbclient, etc.

So QUARTO_PYTHON is effectively QUARTO_JUPYTER in Quarto so might e.g. resolve to this command:

/Users/jjallaire/.julia/conda/3/bin/python3 -m jupyter --paths --json

[dmbates]

Great! Thanks for the clarification.

[jjallaire]

I have just made a change to have Quarto look specifically for the IJulia installed Conda whenever we are using a Julia kernel. This is available in our v1.1.77 pre-release: https://quarto.org/docs/download/prerelease.html.

This should mean that users don't need to worry about having a separate Jupyter installation and should also be more reliable as the IJulia bundled kernel seems less likely to have installation/configuration vagaries that could muck things up.