Jupyter consists of a bunch of small projects, and a few larger ones.
This lays out the basic process of releasing a smaller project,
which should also apply to larger projects,
though they may have some added steps.
Most Jupyter projects use a GitHub milestone system for marking issues and pull requests in releases.
Each release should have a milestone associated with it.
The first step in preparing for a release is to make sure that every issue and pull request has the right milestone.
Go through any open Issues and Pull Requests marked with the current milestone.
If there are any, they need to be resolved or bumped to the next milestone.
It’s fine to bump issues - they are typically marked with the earliest feasible milestone,
but many such optimistically marked tasks aren’t complete when it’s time to release.
There’s always next time!
Check closed Issues and Pull Requests,
using the milestone filter “Issues with no milestone”.
There should never be any closed issues or pull requests without a milestone.
If you find any, go through and mark them with the current milestone or “no action”
A release may be ready to go when it has zero open issues or pull requests.
Once all of the issues and pull requests are dealt with,
it’s time to make release notes.
The smaller projects generally have a changelog.rst in the docs directory,
where you can add a section for the new release.
Look through the pull requests merged for the current milestone (this is why we use milestones),
and write a short summary of the highlights of the changes in this release.
There should generally be a link to the milestone itself for more details.
Make a pull requests with these notes.
It’s a good idea to cc @willingc for review of this PR.
Make sure to mark this PR with your release’s milestone!
Now that your changelog is merged, we can actually build and publish the release.
We’ll assume that V has been declared as a shell variable containing the release version:
Start by making sure you have a clean checkout of master, with no extra files:
git clean -xfd
First, update the version of the package, often in the file <pkg>/_version.py or similar.
Commit that change:
git commit -am "release $V"
At this point, I like to run the tests
just to be sure that setting the version didn’t confuse anything.
Build the distributions:
python setup.py sdist --formats=gztar
python setup.py bdist_wheel
Tag the commit:
git tag -am "release $V" $V
And finally, publish everything, to github and PyPI using twine:
twine upload dist/*
git push origin
git push origin --tags
We have a release! You can now bump the version to the next ‘.dev’ version,
by editing <pkg>/_version.py (or similar) again, and commit:
git commit -am "back to dev"
git push origin
The pushes assume that origin points to the main jupyter/ipython repo.
Depending how you use git, this could be upstream or something else.