Skip to content

Latest commit

 

History

History
278 lines (207 loc) · 11.4 KB

RELEASE.md

File metadata and controls

278 lines (207 loc) · 11.4 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

Using Docker

If desired, you can use Docker to create a new container with a fresh clone of JupyterLab.

First, build a Docker base image. This container is customized with your git commit information. The build is cached so rebuilding it is fast and easy.

docker build -t jlabreleaseimage release/ --build-arg "GIT_AUTHOR_NAME=`git config user.name`" --build-arg "GIT_AUTHOR_EMAIL=`git config user.email`"

Note: if you must rebuild your Docker image from scratch without the cache, you can run the same build command above with --no-cache --pull.

Then run a new instance of this container:

docker rm jlabrelease # delete any old container
docker run -it --name jlabrelease -w /usr/src/app jlabreleaseimage bash

Now you should be at a shell prompt as root inside the docker container (the prompt should be something like root@20dcc0cdc0b4:/usr/src/app).

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 ..

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+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.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 [...packages] x.y.z -> x.y.(z+1) Selected a.b.c -> a.b.(c+1)

Note: if you are making a patch release, and want to update whatever JS packages changed, just do jlpm patch:release js (in fact, any argument, not just js, forces all JS packages to be examined).

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 or rc).

Publishing Packages

Currently we end up with some uncommitted changes at this step. We'll need to commit them before running the publish.

git commit -am "bump version"

Now publish the JS packages and build the python packages

npm run publish:all

If there is a network error during JS publish, run npm run publish:all --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.

At this point, run the source 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, including:

  • committing changes
  • tagging the release
  • and uploading 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
  • 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 0.XX # whatever the new version is
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 path-to-local-extension-cookiecutter-ts
# Fill in the values from the previous branch package.json initial commit
cp -r jupyterlab_apod/ .
rm -rf jupyterlab_apod
  • 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

  • Replace the tag references in the tutorial with the new branch number, e.g. replace 1.0- with 1.1-. Prefix the new tags with the branch name, e.g. 1.0-01-show-a-panel 
    git tag 0.XX-01-show-a-panel HEAD~5
git tag 0.XX-02-show-an-image HEAD~4
git tag 0.XX-03-style-and-attribute HEAD~3
git tag 0.XX-04-refactor-and-refresh HEAD~2
git tag 0.XX-05-restore-panel-state HEAD~1
git tag 0.XX-06-prepare-to-publish HEAD
  • Push the branch with the new tags 
    git push origin 0.XX --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 npm package. Make sure to update the version number in the last commit of the branch. 
    npm publish

If you make a mistake and need to start over, clear the tags using the following pattern:

git tag | grep 0.XX | 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

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.