Mechanism for executing code before running notebook?

[matthew-brett]

In Bookdown, you can specify a before_chapter_script to run code before the notebook (chapter) runs. The typical use-case is setting random number seeds, so the rendered output us predictable, but the notebook itself does not get polluted with seeding code.

I was doing this via some scripting in my own hacked up version of the previous iteration of the Jupyter Book - but is there a way to do this in the current version?

[choldgraf]

Right now it's not possible, though something that has been discussed.

I think we're tracking the idea here: jupyter-book/jupyter-book#777
and sorta related here: executablebooks/MyST-NB#274

The main reason this hasn't been implemented yet is because thus far we've had a fairly strong 1-to-1 mapping of "content file" and "page in the book" to make the contents more self-contained and reproducible. Though this is a feature many have requested.

[matthew-brett]

I may have misunderstood, but what I need in this case is for the code to run only in the case of the rendered notebook, and for the code not to appear in the rendered version or the source, that the student gets by downloading or running the notebook in JupyterHub.

But in general - isn't this a use-at-your-own-risk type of situation? Where we the authors have to be careful we don't put pre-injected code in, that breaks the downloaded notebook, when missing.

[choldgraf]

I'm not personally opposed to the feature, as you say I think it's common across many use-cases. I think many folks want to use this to do things like standardized imports across all of their pages, which would break things. Others have wanted to use it for standardizing things like matplotlib formatting across pages. What I'd personally worry about is that one feature request would lead to another, and if we let people inject code on each page, in a few weeks we'd start seeing issues about why the notebook code and their book outputs don't look the same etc. AKA, "use at your own risk" is reasonable IMO but also might generate maintainer burden and confusion for people regardless.

[akhmerov]

If I were to implement this in my own code, I would do this by using an external script that prepends initialization code to each document. Since myst-nb executes the notebooks extremely early in the pipeline (#144), this would need to happen outside of sphinx.

[matthew-brett]

@choldgraf - honestly I think you do not give users enough credit here. If there is a warning all over this feature, with explanations, I bet you'll find it is used carefully.

[choldgraf]

again I'm not against the feature! Just surfacing points to discuss. I think the place to give feedback like the helpful references to other implementations in R above would be in jupyter-book/jupyter-book#777. If you're interested in trying out a PR to offer up a prototype, I think that the code injection would need to be somewhere before here: https://github.com/executablebooks/MyST-NB/blob/master/myst_nb/parser.py#L74