fix: basic support for bokeh.plotting.show

[akshayka]

Patches bokeh.plotting.show() to display its first argument in the output area. Other arguments are ignored. This is similar to our patch of IPython.display.display.

Not an elegant fix, but should work for many use cases. The proper fix would be to support bokeh's output_notebook() — I believe this uses IPython.display.publish_display_dataunder the hood, which we don't yet support.

cc @bryevdv

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
marimo-docs	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Feb 14, 2025 4:45am
marimo-storybook	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Feb 14, 2025 4:45am

[bryevdv]

@akshayka I'll say I don't have an opinion on changes that you make on your end, if something works for you, that's great. I will state that all of the Bokeh documentation for use in notebooks starts by instructing users to execute output_notebook at the start. You should definitely expect that every Bokeh user will execute output_notebook, so you'd need to at least make sure that calling output_notebook doesn't actively interfere with anything. All that said, I think you could accomplish the same thing without monkeypatching by using Bokehs notebook hook registration function.

[bryevdv]

For reference the default built-in Jupyter hooks are registered like this:

install_notebook_hook('jupyter', load_notebook, show_doc, show_app)

If you used this API you could make your load_notebook (which is what output_notebook does) be a no-op, and you could make show_app emit an exception, since Bokeh server apps can't be run in a WASM notebook. Then show_doc would be your customized version of show.

Edit: I guess you'd also want to set curstate().notebook_type = "marimo" or whatever name you pass to install_notebook_hook so that your hooks get used by default in your notebooks.

[akshayka]

Thanks @bryevdv for the feedback.

That mostly works, except it appears output_notebook() resets curstate().notebook_type to Jupyter, and I don't think I can expect users to manually pass a notebook type to this function (since they'll just copy tutorials like you mentioned).

So it seems like we'd at least need to monkey-patch output_notebook to set curstate().notebook_type to marimo and do nothing else?

[bryevdv]

That mostly works, except it appears output_notebook() resets curstate().notebook_type to Jupyter

If that's true it sounds like a bug or oversight on our end. Do you have a link to the code that you saw doing that handy?

Edit: well, here I guess

https://github.com/bokeh/bokeh/blob/dd4747fa6419fa76f2fe6aa25866457a74e384d6/src/bokeh/io/output.py#L123

So I think there probably out to be a way to override the function parameter default, and it was just an oversight that there is not one.

[akshayka]

So I think there probably out to be a way to override the function parameter default, and it was just an oversight that there is not one.

Thanks, that appears to be it.

For now I think patching show and output_notebook is easiest. I experimented with using the hooks, but it appeared I still needed to call output_notebook(notebook_type="jupyter") for marimo's installed hook to take effect, but that brought in a dependency on IPython that we didn't need.

It's certainly possible I did something wrong, but happy with this simple patch for now.