Drop ipykernel dependency

[SylvainCorlay]

Closes #254.

[blink1073]

We should probably test against both xeus-python and ipykernel on CI.

[echarles]

yep, I was thinking to unit tests that execute cells, so need kernel. +1 to test on both xeus-python and ipykernel on CI although it may sound weird to test something that we wouldn't ship.

[SylvainCorlay]

We should probably test against both xeus-python and ipykernel on CI.

We could do that, although xeus-python is already continuously tested with jupyter_kernel_test.

[blink1073]

Okay, let's test against whichever one is currently shipped with JupyterLab then, and keep them in sync.

[blink1073]

Thanks!

[SylvainCorlay]

Okay, let's test against whichever one is currently shipped with JupyterLab then, and keep them in sync.

We could work on testing a variety of language kernels at some point.

[blink1073]

Julia and R would be good for our namesake. 😀

[kevin-bates]

Good catch - thanks.

[kevin-bates]

Hmm. It just dawned on me. How will the default kernel name be determined? It currently comes from MultiKernelManager.default_kernel_name and defaults to python3 when the trait isn't configured. jupyter_client does not depend on ipykernel either (only for testing), so we could theoretically encounter a condition where the default interaction with kernels now breaks.

[SylvainCorlay]

How will the default kernel name be determined? It currently comes from MultiKernelManager.default_kernel_name and defaults to python3 when the trait isn't configured. jupyter_client does not depend on ipykernel either (only for testing), so we could theoretically encounter a condition where the default interaction with kernels now breaks.

I think that it is fine to default to python3. Erroring when attempting to start a notebook when no kernel is available is expected.

[kevin-bates]

Understood. I'm just saying that previous to this change, there would not have been an error, so there's potential for users claiming a regression. Deployment manifests implicitly relying on that dependency (correctly or not) will need to be updated to include the installation of ipykernel now. That said, switching to jupyter_server is a good time to do this.

[SylvainCorlay]

That said, switching to jupyter_server is a good time to do this.

Yes, that's why I did not open the PR to the notebook server.

We are also removing some of these arbitrary dependencies in ipywidgets 8.0...