python3Packages.numpy: 1.19.4 -> 1.20.1

[rmcgibbo]

Motivation for this change

https://numpy.org/devdocs/release/1.20.0-notes.html

Things done

• Tested using sandboxing (nix.useSandbox on NixOS, or option sandbox in nix.conf on non-NixOS linux)
• Built on platform(s)
  • NixOS
  • macOS
  • other Linux distributions
• Tested via one or more NixOS test(s) if existing and applicable for the change (look inside nixos/tests)
• Tested compilation of all pkgs that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review wip"
• Tested execution of all binary files (usually in ./result/bin/)
• Determined the impact on package closure size (by running nix path-info -S before and after)
• Ensured that relevant documentation is up to date
• Fits CONTRIBUTING.md.

---

Based on joblib/joblib#1136 (and my testing), there's an incompatibility between numpy 1.20 and joblib <= 1.0, so I've also bumped joblib in this PR as well.

I did not rebuild everything since there are quite a lot of downstream packages that depend on numpy, but I did to a random selection of 62 packages:

NIXPKGS_ALLOW_BROKEN=1 nix build -f .  python39Packages.pyrr python39Packages.pysptk python39Packages.pysqueezebox python39Packages.pysrim python39Packages.pyssim python39Packages.pystray python39Packages.pytesseract python39Packages.pytest-aiohttp python39Packages.pytest-arraydiff python39Packages.pytest-black python39Packages.pytest-doctestplus python39Packages.pytest-filter-subpackage python39Packages.pytest-httpbin python39Packages.pytest-mpl python39Packages.pytest-randomly python39Packages.pytest-sanic python39Packages.pytest-vcr python39Packages.python-engineio python39Packages.python-miio python39Packages.python-mystrom python39Packages.python-opendata-transport python39Packages.python-socketio python39Packages.pytools python39Packages.pytorch python39Packages.pytorch-lightning python39Packages.pytorch-metric-learning python39Packages.pyuavcan python39Packages.pywbem python39Packages.pyworld python39Packages.pyxeoma python39Packages.py_stringmatching python39Packages.qdldl python39Packages.qimage2ndarray python39Packages.qrcode python39Packages.quantities python39Packages.rainbowstream python39Packages.rasterio python39Packages.regenmaschine python39Packages.regional python39Packages.reikna python39Packages.reportlab python39Packages.resampy python39Packages.ripser python39Packages.rising python39Packages.roboschool python39Packages.rtmixer python39Packages.s3fs python39Packages.sasmodels python39Packages.scapy python39Packages.scikit-fmm python39Packages.scikit-optimize python39Packages.scikit-tda python39Packages.scipy python39Packages.scrapy-deltafetch python39Packages.scrapy-fake-useragent python39Packages.scrapy-splash python39Packages.scs python39Packages.seabreeze python39Packages.seekpath python39Packages.seqdiag python39Packages.shapely python39Packages.sharedmem --keep-going

The failures are:

• python39Packages.rainbowstream: Also broken currently on hydra staging.
• python39Packages.fixtures. Also broken on currently on hydra staging.
• python39Packages.sasmodels: Also broken currently on hydra staging.
• python39Packages.scikitlearn: Not sure why. I think these are new failures not on staging though.

This is the scikit-learn failure: https://gist.github.com/rmcgibbo/21392d1c75d2c0809533bc3a97dce4ba

[rmcgibbo]

Based on this comment, on the scikit-learn tracker, I believe that scikit-learn crash is reproduced on their CI and also not yet fixed, so I don't think there's a patch we can pull in yet.

[dotlambda]

This should target staging, not staging-next.

[dotlambda]

Both commit messages should start with python3Packages.

[rmcgibbo]

It's probably best not to break scikit-learn, since that package has a lot of reverse dependencies.

Therefore, I think either I should disable the failing scikit-learn test, or we should wait to merge this PR until a patch is available from scikit-learn upstream, whichever ya'll think is best.

[rmcgibbo]

Therefore, I think either I should disable the failing scikit-learn test, or we should wait to merge this PR until a patch is available from scikit-learn upstream, whichever ya'll think is best.

I found the scikit-learn patch that makes it work with numpy 1.20.0

[dotlambda]

6 packages built:
python37Packages.matplotlib python37Packages.scikitlearn python38Packages.matplotlib python38Packages.scikitlearn python39Packages.matplotlib python39Packages.scikitlearn

[rmcgibbo]

Let's not merge this before I update the PR to 1.20.1 instead: https://mail.python.org/pipermail/scipy-dev/2021-February/024523.html