List of out of date packages

[hoodmane]

The following pure Python packages with patches are out of date

• astropy v3.2.3 ==> 4.2.1
• joblib 0.11 ==> 1.0.1
• mpmath 1.1.0 ==> 1.2.1
• numcodecs 0.7.2 ==> 0.7.3
• pytest 3.6.3 ==> 6.2.3 Updated by Update pytest #1719.
• python-sat 0.1.6.dev6 ==> 0.1.7.dev1
• scikit-image 0.15.0 ==> 0.18.1 : blocked by scipy
• setuptools 40.0.0 ==> 56.0.0
• zarr 2.6.1 ==> 2.7.0

Following mixed packages are out of date:

• numpy 1.17.5 ==> 1.20.2
• libxml 4.4.1 ==> 4.6.3
• pillow 8.0.1 ==> 8.2.0
• pandas 1.0.5 ==> 1.2.4
• matplotlib 3.3.3 ==> 3.4.1
• scipy 0.17.1 ==> 1.6.2 : WIP Update to scipy 1.6.1 #1293
• scikit-learn 0.22.2 ==> 0.24.1 : blocked by scipy

[teonbrooks]

mne is out of date as well but I've been thinking about whether it's necessary to keep it in the package list since it is a pure python package.
I'm using pyodide in an Electron app and it's been convenient having mne bundled with the zip but I was thinking it might be useful to download the pypi wheel, and install it via micropip (I need the app to be able to run without internet so I can't load mne on the fly). It would help reduce the number of extraneous packages.
just to confirm, can you load wheels locally? if so, would all i need to do is load pyodide in JS and do micropip.install('./path/to/mne/mne_something.whl')?

[hoodmane]

mne will be updated to 0.22.1 in #1460.

just to confirm, can you load wheels locally?

I believe that this works and we intend to support it. But if it turns out that it doesn't work for some reason, please open an issue. =)

[teonbrooks]

ok cool :) I think I will wait til after the release to propose removing it. at the moment, it's harmless. just curious, is there a guiding principle to what gets added/removed to pyodide internal package manager? I assume requested non-pure py packages. this all might change if we worked to get conda(-forge) to make a wasm build target :)

[hoodmane]

is there a guiding principle to what gets added/removed to pyodide internal package manager? I assume requested non-pure py packages

Yeah I think the policy is something like: we won't take anything that has a pure Python wheel on pypi.

If it is pure Python and has no wheel, please make a reasonable effort to get the maintainers of the package to add a wheel, if that doesn't work in a reasonable amount of time then we can add the package until the wheel gets added. See the case with plotly / retrying in #1267.

If it is not pure Python, then we get into a cost/benefit analysis of build time and maintenance effort vs usefulness / interest.

I suppose I should make a PR to write this down in the docs.

See also: #1397

[hoodmane]

I also want to combine micropip and loadPackage and rework things to make it matter less whether a package is loaded from a wheel or not. Currently the API is a bit hard to use and makes it a breaking change to remove a pure Python package. See #1470.

[hoodmane]

@teonbrooks mne raises an error on import: ValueError: LAPACK function dgesvd could not be found
since version 0.19.0. No update for mne I guess.

[rth]

mne raises an error on import: ValueError: LAPACK function dgesvd could not be found
since version 0.19.0. No update for mne I guess.

Thanks for checking. So it also needs a newer scipy.

[hoodmane]

Closed as outdated. We currently down to 8 out of date packages after #2085.