anaconda / nb_conda_kernels

Package for managing conda environment-based kernels inside of Jupyter
BSD 3-Clause "New" or "Revised" License
601 stars 70 forks source link

Two-arch builds #244

Closed mcg1969 closed 8 months ago

mcg1969 commented 8 months ago

We can't make this a true noarch package because of the pywin32 dependency. (conda-forge can do so because they have the magical pywin32-on-windows package that Anaconda does not support. That may change, but not now.)

However, what Anaconda does support are virtual packages like __unix and __win. This would allow us to build two noarch packages in parallel that cover all Python versions.

mathause commented 8 months ago

See https://github.com/conda-forge/nb_conda_kernels-feedstock/pull/52?

mcg1969 commented 8 months ago

Excellent, thanks. I'll work on bringing that over here. Note that I'm thinking of this more from the perspective of testing; I want to be able to test the package in as close to its built form as possible The fact that we have noarch packages on conda-forge but we're building them arch over here is a discrepancy I don't like.

mathause commented 8 months ago

The noarch is relatively recent (conda-forge/nb_conda_kernels-feedstock#50) but the post init scripts were not called on windows, so I fixed that with the virtual packages (as suggested by ocepaf)

mcg1969 commented 8 months ago

That raises another issue... I believe the scripts themselves can potentially be retired, at least on new-ish versions of notebook and/or jupyter_server.