Skip to content

Latest commit

 

History

History
282 lines (210 loc) · 11.2 KB

RELEASE.md

File metadata and controls

282 lines (210 loc) · 11.2 KB
Raw

Making a JupyterLab release

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

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 minor x.y.z-> x.(y+1).0.a0 All a.b.c -> a.(b+10).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.a1-> x.y.z.rc0 All a.b.c-alpha.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 patch:release x.y.z -> x.y.(z+1) Changed a.b.c -> a.b.(c+1)

Note: For a minor 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]

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 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.