Skip to content

Latest commit

 

History

History
476 lines (330 loc) · 20 KB

RELEASE.md

File metadata and controls

476 lines (330 loc) · 20 KB
Raw

Making a JupyterLab release

This document guides a contributor through creating a release of JupyterLab.

The JupyterLab Release Process

JupyterLab follows semver for the versioning of the Python and JavaScript packages.

Release Timeline

Although the commitments listed below are "best effort", the JupyterLab team tries to follow a couple of guidelines:

  • one major version per year, which usually includes API breaking changes
  • several minor versions per year that include new features but no API breaking changes
  • support and bug fixes on a couple of final releases (by backporting PRs and releasing from release branches)

Release Plans are tracked in dedicated issues, and are closed when the final release. See the following two issues as an example:

Alpha Releases

Alpha releases have a fairly low bar. Their purpose is to start putting the new JupyterLab version into the hands of users and extension authors.

The requirements for an alpha release should be that JupyterLab can be installed and run. Bugs and breaking changes are accepted.

Beta Releases

Beta releases usually try to not have breaking changes in the API, although breaking changes can sometimes happen during that phase if they were missed during the alpha stage.

The recommended time period for the Beta phase is a minimum of 2 weeks.

The draft changelog describing user-facing changes will be published with the first Beta release.

The community of extension developers and active users will be invited to commence testing the new Beta release including the draft user-facing changelog, and an invitation to open issues for any major:

  • regressions,
  • usability problems
  • points needing clarification (or inclusion) in the changelog, and
  • points needing clarification in the extension porting guide.

The start of the Beta-testing period will be announced on Jupyter mailing group and Jupyter Discourse for major releases, and only via a Discourse post for minor releases.

All bug reports raised during the Beta-testing period should be triaged (but not necessarily addressed) before releasing the first release candidate.

Release Candidates

Release Candidates (RC) are a signal to the extension developer community that they should start migrating to the new version to test it. At that point we consider the software stable.

The RC stage is often a good time to address final release documentation changes or minor UX tweaks. During the RC phase, the JupyterLab developers and maintainers start updating third-party extensions over to the new version to test it. This work during the RC phase, and giving time for feedback from extension developers, can take up to a couple of weeks.

The recommended time period for the Release Candidate phase is a minimum of 1 week for minor releases, and 2 weeks for major releases.

Automated Releases with the Jupyter Releaser

The recommended way to make a release is to use jupyter_releaser.

Workflow

The full process is documented in https://jupyter-releaser.readthedocs.io/en/latest/get_started/making_first_release.html#making-your-first-release. There is a recording of the full workflow on YouTube.

Here is a quick summary of the different steps.

Communicate on Gitter

It is good practice to let other maintainers and users know when starting a new release.

For this we usually leave a small message in the jupyterlab room on Gitter: https://gitter.im/jupyterlab/jupyterlab. Once the release is done, we also post a message with a link to the release notes, which include the changelog.

Draft Changelog

The first step is to generate a new changelog entry for the upcoming release.

We use the "Draft Changelog" workflow as documented here: https://jupyter-releaser.readthedocs.io/en/latest/get_started/making_first_release.html#draft-changelog

The workflow takes a couple of input parameters. Here is an overview with example values:

Input Description Example Value
Target The owner/repo GitHub target jupyterlab/jupyterlab
Branch The branch to target master
Version Spec New Version Spec next
Since Use PRs since activity since this date or git reference v4.0.0a15

The version spec follows the specification documented below in the Bump Version section.

We can use next when making a patch release or a build pre-release.

Click on "Run workflow", then wait for:

  1. the PR to be created on the repo. Example: jupyterlab#11422
  2. Tests to pass
  3. Merge the changelog PR

Full Release

PyPI and npm tokens

Before running the "Full Release" workflow, make sure you have been added to:

Then create the PyPI and npm tokens. Check out the links in the Jupyter Releaser Setup Documentation for more information.

Running the workflow

On the jupyter_releaser fork, select the "Full Release" workflow.

Fill in the information as mentioned in the body of the changelog PR, for example:

Input Value
Target jupyterlab/jupyterlab
Branch master
Version Spec next
Since v4.0.0a15

The "Full Release" workflow:

  • builds and uploads the jupyterlab Python package to PyPI
  • builds the @jupyterlab/* packages and uploads them to npm
  • creates a new GitHub Release with the new changelog entry as release notes
  • creates a PR to forward port the new changelog entry to the main branch (when releasing from a branch that is not the default)

Then follow the Post release candidate checklist if applicable.

Manual Release Process

Review CONTRIBUTING.md. Make sure all the tools needed to generate the built JavaScript files are properly installed.

Creating a full release

We publish the npm packages, a Python source package, and a Python universal binary wheel. We also publish a conda package on conda-forge (see below). See the Python docs on package uploading for twine setup instructions and for why twine is the recommended method.

Getting a clean environment

For convenience, here is a script for getting a completely clean repo. This makes sure that we don't have any extra tags or commits in our repo (especially since we will push our tags later in the process), and that we are on the correct branch. The script creates a conda env, pulls down a git checkout with the appropriate branch, and installs JupyterLab with pip install -e ..

Make sure you are running an sh-compatible shell, and it is set up to be able to do conda activate. Then do:

source scripts/release_prep.sh <branch_name>

Bump version

The next step is to bump the appropriate version numbers. We use bump2version to manage the Python version, and we keep the JS versions and tags in sync with the release cycle.

Here is an example of how version numbers progress through a release process. Choose and run an appropriate command to bump version numbers for this release.

Command Python Version Change NPM Version change
jlpm bumpversion major x.y.z-> (x+1).0.0.a0 All a.b.c -> a.(b+10).0-alpha.0
jlpm bumpversion minor x.y.z-> x.(y+1).0.a0 All a.b.c -> a.(b+1).0-alpha.0
jlpm bumpversion build x.y.z.a0-> x.y.z.a1 All a.b.c-alpha.0 -> a.b.c-alpha.1
jlpm bumpversion release x.y.z.a1-> x.y.z.b0 All a.b.c-alpha.1 -> a.b.c-beta.0
jlpm bumpversion release x.y.z.b1-> x.y.z.rc0 All a.b.c-beta.1 -> a.b.c-rc.0
jlpm bumpversion release x.y.z.rc0-> x.y.z All a.b.c-rc0 -> a.b.c
jlpm bumpversion patch x.y.z -> x.y.(z+1) Changed a.b.c -> a.b.(c+1)

Note: For a major release, we bump the JS packages by 10 versions so that we are not competing amongst the minor releases for version numbers. We are essentially sub-dividing semver to allow us to bump minor versions of the JS packages as many times as we need to for minor releases of the top level JupyterLab application.

Other note: It's ok if yarn-deduplicate exits with a non zero code. This is expected!

JS major release(s)

In a major Python release, we can have one or more JavaScript packages also have a major bump. During the prerelease stage of a major release, if there is a backwards-incompatible change to a JS package, bump the major version number for that JS package:

jlpm bump:js:major [...packages]

NOTE You should rebase before running jlpm bump:js:major to avoid a cascade of merge conflicts.

Results:

  • Python package is not affected.
  • JS dependencies are also bumped a major version.
  • Packages that have already had a major bump in this prerelease cycle are not affected.
  • All affected packages changed to match the current release type of the Python package (alpha, beta, or rc).

Publishing Packages

Now publish the JS packages

npm run publish:js

If there is a network error during JS publish, run npm run publish:js --skip-build to resume publish without requiring another clean and build phase of the JS packages.

Note that the use of npm instead of jlpm is significant on Windows.

Next, prepare the python release by running:

npm run prepare:python-release

This will update the Python package to use the new JS packages and create the Python release assets. Note: sometimes the npm registry is slow to update with the new packages, so this script tries to fetch the packages until they are available.

At this point, run the ./scripts/release_test.sh to test the wheel in a fresh conda environment with and without extensions installed. Open and run the Outputs notebook and verify everything runs properly. Also add a cell with the following code and make sure the widget renders:

from ipywidgets import IntSlider
IntSlider()

Finish

Follow instructions printed at the end of the publish step above:

twine upload dist/*
git push origin --tags <BRANCH>

These lines:

  • upload to pypi with twine
  • double-check what branch you are on, then push changes to the correct upstream branch with the --tags option.

Post release candidate checklist

Now do the actual final release:

  • Run jlpm run bumpversion release to switch to final release
  • Push the commit and tags to master
  • Run npm run publish:all to publish the packages
  • Create a branch for the release and push to GitHub
  • Update the API docs
  • Merge the PRs on the other repos and set the default branch of the xckd repo
  • Publish to conda-forge.

After a few days (to allow for possible patch releases), set up development for the next release:

  • Run jlpm run bumpversion minor to bump to alpha for the next alpha release
  • Put the commit and tags to master
  • Run npm run publish:all to publish the packages
  • Release the other repos as appropriate
  • Update version for binder

Updating the extension tutorial

  • Clone the repo if you don't have it
git clone git@github.com:jupyterlab/jupyterlab_apod.git

Simple updates by rebasing

If the updates are simple, it may be enough to check out a new branch based on the current base branch, then rebase from the root commit, editing the root commit and other commits that involve installing packages to update to the new versions:

git checkout -b BRANCH # whatever the new version is, e.g., 1.0
git rebase -i --root

"Edit" the commits that involve installing packages, so you can update the package.json. Amend the last commit to bump the version number in package.json in preparation for publishing to npm. Then skip down to the step below about publishing the extension tutorial. If the edits are more substantial than just updating package versions, then do the next steps instead.

Creating the tutorial from scratch

  • Create a new empty branch in the extension repo.
git checkout --orphan name-of-branch
git rm -rf .
git clean -dfx
cookiecutter -o initial path-to-local-extension-cookiecutter-ts
# Fill in the values from the previous branch package.json initial commit
cp -r initial/jupyterlab_apod .
rm -rf initial
  • Create a new PR in JupyterLab.
  • Run through the tutorial in the PR, making commits and updating the tutorial as appropriate.
  • For the publish section of the readme, use the README file from the previous branch, as well as the package.json fields up to license. Bump the version number in preparation for publishing to npm.

Publishing extension tutorial changes

  • Tag commits in the branch with the appropriate branch-step tag. If you are at the final commit, you can tag all commits with the below, replacing BRANCH with the branch name (e.g., 1.0-01-show-a-panel) ```bash git tag BRANCH-01-show-a-panel HEAD4 git tag BRANCH-02-show-an-image HEAD3 git tag BRANCH-03-style-and-attribute HEAD2 git tag BRANCH-04-refactor-and-refresh HEAD1 git tag BRANCH-05-restore-panel-state HEAD

  • Push the branch with the new tags

    git push origin BRANCH --tags

    Set the branch as the default branch (see github.com/jupyterlab/jupyterlab_apod/settings/branches).

  • If there were changes to the example in the documentation, submit a PR to JupyterLab

  • Publish the new jupyterlab_apod python package. Make sure to update the version number in the last commit of the branch.

    twine upload dist/*

If you make a mistake and need to start over, clear the tags using the following pattern (replacing BRANCH with the branch name):

git tag | grep BRANCH | xargs git tag -d

Publishing to conda-forge

  • If no requirements have changed, wait for the conda-forge autotick-bot.
  • Otherwise:
  • Get the sha256 hash for conda-forge release:
shasum -a 256 dist/*.tar.gz

Updating API Docs

Run source scripts/docs_push.sh to update the gh-pages branch that backs http://jupyterlab.github.io/jupyterlab/.

Making a manual patch release

  • Backport the change to the previous release branch
  • Run the following script, where the package is in /packages/package-folder-name (note that multiple packages can be given, or no packages for a Python-only patch release):
jlpm run patch:release package-folder-name
  • Push the resulting commit and tag

Update version for binder

Each time we release JupyterLab, we should update the version of JupyterLab used in binder and repo2docker. Here is an example PR that updates the relevant files:

https://github.com/jupyter/repo2docker/pull/169/files

This needs to be done in both the conda and pip buildpacks in both the frozen and non-frozen version of the files.

Making a Minor Release

Planning

  • Create a pinned issue
  • Create a milestone
  • Decide on a scope for the release and set a target final release date

Alpha and Beta Phase

  • Create a new branch from the previous release branch
  • Use a ".x" in the branch name so we can continue to use it for patches
  • Update branch and RTD config in ensure_repo.ts and run jlpm integrity to update links - source should be the previous release branch
  • Update readthedocs branch config as appropriate
  • Automated Release using "minor" - edit changelog for new section
  • Move through alpha and beta phases as appropriate

RC Phase

  • Roll up the release notes using the "Use PRs with activity since the last stable git tag" option when running the workflows
  • Update the release issue with an updated date

Final Release

  • Roll up the release notes using the "Use PRs with activity since the last stable git tag" option when running the workflows
  • Close the release issue
  • Rename milestone to use ".x"
  • Make an announcement on Discourse

Making a Major Release

Planning

  • Create a pinned issue
  • Create a milestone
  • Decide on a scope for the release and set a target final release date

Alpha and Beta Phase

  • Update branch and RTD config in ensure_repo.ts and jlpm integrity to update links - source should be the previous branch
  • Update readthedocs branch config as appropriate
  • Automated Release using "major" - edit changelog for new section
  • Move through alpha and beta phases as appropriate

RC Phase

  • Roll up the release notes using the "Use PRs with activity since the last stable git tag" option when running the workflows
  • Create a new branch from the default branch with ".x" in the name so we can continue to use it for patches
  • Update the release issue with an updated date

Final Release

  • Roll up the release notes using the "Use PRs with activity since the last stable git tag" option when running the workflows
  • Close the release issue and rename milestone to use ".x"
  • Make an announcement on Discourse
  • Make a blog post

Postmortems

Here is a list of previous issues that happened while releasing JupyterLab, that can be used as reference in case new issues show up in the future: