orchest / orchest

Describe the feature you'd love to have in Orchest

It would be nice for Orchest to install ipywidgets on Jupyter by default.

Is your feature request related to a problem? Please describe.

Initially I was not able to make anything based on ipywidgets work, but then I realized I probably needed to install it globally for Jupyter, and not as part of a particular environment. That turned out to be the solution.

However, I don't think our average user will reach that conclusion so quickly.

Additional context

One alternative is just to document this better: for example, by adding a "Error displaying widget: model not found" entry to the FAQ. But both approaches are not mutually exclusive: we can add that entry to the FAQ anyway, and still install ipywidgets by default, which is quite common.

I remember implementing this back in the day, don't the following lines install ipywidgets for the user?

orchest/services/jupyter-server/requirements.in

Line 4 in 89d5bdc

jupyterlab-widgets

and

orchest/services/base-images/runnable-shared/runner/requirements-user.txt

Line 2 in b90c2e1

ipywidgets

Sounds also like the approach that you took?

Looks like it, but at the same time doesn't seem to work.

To reproduce:

Verify that the JupyterLab Configuration script is empty
Create a new project and a new pipeline
~~Add pip install ipywidgets in the setup script~~ They are being preinstalled already
Get to JupyterLab and do import ipywidgets as ipyw; ipyw.IntSlider()
See Error displaying widget: model not found ❌
Go back to JupyterLab configuration, add pip install ipywidgets
Repeat import ipywidgets as ipyw; ipyw.IntSlider(), see the slider ✔️

Note: requirements-user.txt installs ipywidgets inside the environment (my step 4 from above, insufficient 👎🏽) and it doesn't seem like jupyterlab-widgets depends on ipywidgets.

I acknowledge that this procedure follows the official documentation though: https://ipywidgets.readthedocs.io/en/stable/user_install.html#installing-in-jupyterlab-3-x

I opened jupyter-widgets/ipywidgets#3577 to track this seemingly unexpected behavior upstream.

Confirmed that upgrading jupyterlab_widgets on the JupyterLab environment to 3.0 worked.

Mistery solved, PR incoming.

I'm noting though that this might cause some incompatibilities with ipympl... matplotlib/ipympl#460 (comment)

Repurposed gh-1242 to cap ipywidgets, instead of upgrading it. This still solves the issue while giving time for the ecosystem to adapt.

ipywidgets gives `Model not found` errors