Move CI to GitHub Actions #1074
Conversation
setup.py: update trove classifiers setup.py: use setuptools.find_packages() setup.py: use version from git tag Note to self: git-tag releases since 0.7.3 !!! rm MANIFEST.in since handled by setuptools_scm MNT: setup.py: add dev deps Remove pyproject.toml and requirements.txt We already have more versatile setup.py + setup.cfg Add docs dependencies to setup.py setup.py: Add project_urls, shown in PyPI sidebar setup.py: Use SPDX license identifier Remove doc/requirements.txt in favor of setup.extras_require setup.py: add pytest-cov dev dependency CI: Require pytest-xdist for 'test-code-parallel' make target
Add .github/PULL_REQUEST_TEMPLATE.md
Drop CI testing on python 3.6 -- it's EOL by end of the year. Update README.rst CI badge DOC: Fix error "Latexmk: Missing input file: 'tgtermes.sty' from line" CI: Install in develop mode for Docs Since `make -C doc doctest` is run from the project dir, sphinx/Python may import the curdir skopt instead of the installed one. Install in develop mode to get skopt/_version.py file created. The alternative would be to cd elsewhere before running make.
Remove media dir; use intro image from doc/image/bo-objective.png DOC: Merge and extend README/Development into CONTRIBUTING.md Point docs/development to CONTRIBUTING.md as the canonical source
To v0.22.* -- it was released in 2019. Also this joblib/joblib#917 (comment) Futher bump tested scikit-learn == 0.24.* -- it is old enough! The alternative would be to pin equally ancient scipy to avoid error: > AttributeError: 'str' object has no attribute 'decode'
The test was "rediscovered" (in a failing state) by renaming the test case that follows it.
|
Hello @kernc! Thanks for updating this PR. We checked the lines you've touched for PEP聽8 issues, and found:
Comment last updated at 2021-10-11 20:15:31 UTC |
.github/workflows/ci.yml
Outdated
| - name: Publish 馃摝 on PyPI | ||
| if: env.DO_PUBLISH == 'true' | ||
| uses: pypa/gh-action-pypi-publish@master | ||
| with: | ||
| password: ${{ secrets.PYPI_API_TOKEN }} |
There was a problem hiding this comment.
Eventually, this secret would need to be set to push new packages to PyPI.
.github/workflows/ci.yml
Outdated
| - name: Publish docs | ||
| if: env.DO_PUBLISH_DOCS == 'true' | ||
| env: | ||
| GITHUB_TOKEN: ${{ secrets.GITHUB_DOCS_TOKEN }} | ||
| run: | | ||
| .github/scripts/publish_docs.sh doc/_build/html/stable |
There was a problem hiding this comment.
And this secret would need to be set to push generated docs to the docs repo.
|
@pep8speaks in pep8speaks-org/pep8speaks#95 says it should follow setup.cfg flake8 configuration, but apparently not the case? I'd kindly motion toggling off the noise in favor of a more simple lint check. 馃憤 |
| @@ -0,0 +1,161 @@ | |||
| name: CI | |||
There was a problem hiding this comment.
As I mentioned in Issue #1060, you're going to want to factorize the CI into separate workflows. GHA was not designed to have everything be shoved into a single config like Travis or Circle.
There was a problem hiding this comment.
I was looking into it, but I found no way to properly DRY/reuse build steps in the publish workflow. They just now made available reusable workflows. Unless you have a better idea for it, I'll look into that. 馃憤
There was a problem hiding this comment.
The reusable workflows are indeed nice, but they are more useful at an org level (though still can be used at the project level if really desired). GHA still are designed to be used in a more modular fashion than Travis or Circle.
|
Should scikit-optimize leverage this PR to move to the black code-style, like scikit-learn did ? |
Fixes #1060
Closes #1072
Closes #1068
Closes #1069
Fixes #1073
This took a while, but it's often better late than never. 馃槄
The PR looks huge, but it's namely about:
I tried my best to simplify, yet continue @holgern's work鈥攚hat happened, man, have you died?! :D
It's not too pretty, and it's opinionated in some places, but it's what I can offer and am willing to maintain. The commits have been roughly squashed by function. The PR is at this point certainly ready for review. You can see all the checks passing on a PR to my fork, here.