How to release plotly packages

There are 3 Python packages (plotly, plotly-geo and chart-studio) which need to be published to PyPI and conda, and 1 JS packages (jupyterlab-plotly) which need to be published to NPM. In addition, there are various changelogs, github releases and forum announcements to do :)

Release process - `plotly` package and extensions

This is the release process for releasing plotly.py version X.Y.Z with jupyterlab-plotly with matching versions.

Note: it's easier to lock the JS extension and Python versions together, even if it means we occasionally push no-change versions to NPM/PyPI/Conda.

Finalize changelog

Review the contents of packages/python/plotly/CHANGELOG.md. We try to follow the keepachangelog guidelines. Make sure the changelog includes the version being published at the top, along with the expected publication date.

Use the Added, Changed, Deprecated, Removed, Fixed, and Security labels for all changes to plotly.py. If the version of plotly.js has been updated, include this as the first Updated entry. Call out any notable changes as sub-bullets (new trace types in particular), and provide a link to the plotly.js CHANGELOG.

Finalize versions

Manually update the versions to X.Y.Z in the files specified below.

CHANGELOG.md
- update the release date
packages/python/plotly/README.md
- this must be done at this point because the README gets baked into PyPI
plotly/_widget_version.py:
- Update __frontend_version__ to ^X.Y.Z (Note the ^ prefix)
packages/javascript/jupyterlab-plotly/package.json
- Update "version" to X.Y.Z
- Ensure you're using node version 12 and npm version 6 to minimize diffs to package-lock.json
- Ensure you're in a Python virtual environment with JupyterLab 3 installed
- Run rm -rf node_modules && npm install && npm run build:prod
This the last good time to install the extensions locally and check that everything works in dev mode
Run git diff and ensure that only the files you modified and the build artifacts have changed
Ensure that the diff in package-lock.json seems sane
Commit and tag but don't push until after everything is available on NPM/PyPI/Conda (see below):
- git commit -a -m "release vX.Y.Z"
- git tag vX.Y.Z

Publish JS Extensions to NPM

Build and publish the final version of the extensions to NPM. We do this first because once we push to PyPI the README will refer to these versions.

cd packages/javascript/jupyterlab-plotly
npm run build && npm publish --access public

Final checks could be done here if desired.

Publishing to PyPI

Build and publish the final version to PyPI.

(plotly_dev) $ git status # make sure it's not dirty!
(plotly_dev) $ cd packages/python/plotly
(plotly_dev) $ rm -rf dist
(plotly_dev) $ python setup.py sdist bdist_wheel
(plotly_dev) $ rm -f dist/*dirty*
(plotly_dev) $ twine upload dist/plotly-X.Y.Z*

Note: this will intentionally fail if your current git tree is dirty, because we want the tag to reflect what is being released, and the version number comes from the tag and the dirty-state.

After it has uploaded, move to another environment and double+triple check that you are able to upgrade ok:

$ pip install plotly --upgrade

And ask one of your friends to do it too. Our tests should catch any issues, but you never know.

Publishing to the plotly conda channel

To publish package to the plotly anaconda channel you'll need to have the anaconda or miniconda distribution installed, and you'll need to have the anaconda-client package installed.

(plotly_dev) $ conda config --set anaconda_upload no
(plotly_dev) $ conda build recipe/

Then upload artifacts to the anaconda channel by running the upload command that conda provides, which looks something like this:

$ anaconda upload /path/to/anaconda3/conda-bld/noarch/plotly-*.tar.bz2

Push the commit and add GitHub Release entry

(plotly_dev) $ git push origin master
(plotly_dev) $ git push origin vX.Y.Z

Go to https://github.com/plotly/plotly.py/releases and "Draft a new release"
Enter the vX.Y.Z tag you created already above and make "Release title" the same string as the tag.
Copy the changelog section for this version as the "Describe this release"

Update documentation site

Search for the previous version string in the docs and replace it with the new version string, including but not necessarily limited to the following files:
- doc/python/getting-started.md
- doc/apidoc/conf.py
- doc/requirements.txt
- binder/requirements.txt
doc-prod should already have been merged on a regular basis into master, but start by doing it first if not. Then merge master into doc-prod to deploy the doc related to features in the release.
in a clone of the graphing-library-docs repo:
1. bump the version of Plotly.js with cd _data && python get_plotschema.py <PLOTLY.JS VERSION> fixing any errors that come up
2. rebuild the Algolia schema index with ALGOLIA_API_KEY=<key> make update_ref_search
3. Rebuild the Algolia python index with ALGOLIA_API_KEY=<key> make update_python_search
4. Commit and push the changes to master in that repo

Notify Stakeholders

Post an announcement to the Plotly Python forum, with links to the README installation instructions and to the CHANGELOG.
Update the previous announcement to point to this one
Update the Github Release entry and CHANGELOG entry to have the nice title and a link to the announcement
Follow up on issues resolved in this release or forum posts with better answers as of this release

Release Candidate process - `plotly` package

Bump to release candidate version

Manually update the versions to X.Y.Z-rc.1 in the files specified below.

packages/python/plotly/plotly/_widget_version.py:
- Update __frontend_version__ to ^X.Y.Z-rc.1 (Note the ^ prefix)
packages/javascript/jupyterlab-plotly/package.json
- Update "version" to X.Y.Z-rc.1
- Ensure you're using node version 12 and npm version 6 to minimize diffs to package-lock.json
- Ensure you're in a Python virtual environment with JupyterLab 3 installed
- Run rm -rf node_modules && npm install && npm run build:prod

Commit the changes
Tag this commit on the release branch as vX.Y.Zrc1

In both cases rc is the semantic versioning code for Release Candidate.

The number 1 means that this is the first release candidate, this number can be incremented if we need to publish multiple release candidates. Note that the npm suffix is -rc.1 and the PyPI suffix is rc1.

Publishing plotly.py and jupyterlab-plotly as release candidates allows us to go through the publication process, and test that the installed packages work properly before general users will get them by default. It also gives us the opportunity to ask specific users to test that their bug reports are in fact resolved before we pull the trigger on the official release.

Publish release candidate to PyPI

To upload to PyPI you'll also need to have twine installed:

(plotly_dev) $ pip install twine

And, you'll need to be a maintainer on PyPI. Then, from inside the repository:

(plotly_dev) $ cd packages/python/plotly
(plotly_dev) $ git checkout release_X.Y.Z
(plotly_dev) $ git stash
(plotly_dev) $ rm -rf dist
(plotly_dev) $ python setup.py sdist bdist_wheel
(plotly_dev) $ rm -f dist/*dirty*
(plotly_dev) $ twine upload dist/plotly-X.Y.Zrc1*

Note: this will intentionally fail if your current git tree is dirty, because we want the tag to reflect what is being released, and the version number comes from the tag and the dirty-state.

Publish release candidate of JS Extensions to NPM

Now, publish the release candidate of the extensions to NPM.

cd ./packages/javascript/jupyterlab-plotly
npm run build && npm publish --access public --tag next

The --tag next part ensures that users won't install this version unless they explicitly ask for the version or for the version with the next tag.

Publish release candidate to plotly anaconda channel

To publish package to the plotly anaconda channel you'll need to have the anaconda or miniconda distribution installed, and you'll need to have the anaconda-client package installed.

(plotly_dev) $ conda config --set anaconda_upload no
(plotly_dev) $ conda build recipe/

Next run anaconda login and enter the credentials for the plotly anaconda channel.

Then upload artifacts to the anaconda channel using the test label. Using the test label will ensure that people will only download the release candidate version if they explicitly request it.

$ anaconda upload --label test /path/to/anaconda3/conda-bld/noarch/plotly-*.tar.bz2

Then logout with anaconda logout

Manually test the release candidate

Create a fresh virtual environment (or conda environment) and install the release candidate by following the new README.md instructions (the instructions updated above to include the release candidate versions)

Run through the example notebooks at https://github.com/jonmmease/plotly_ipywidget_notebooks using the classic notebook and JupyterLab. Make sure FigureWidget objects are displayed as plotly figures, and make sure the in-place updates and callbacks work.

If appropriate, ask users who have submitted bug reports or feature requests that are resolved in this version to try out the release candidate.

If problems are found in the release candidate, fix them on the release branch and then publish another release candidate with the candidate number incremented.

Release process - `plotly-geo` package

The plotly-geo package contains the shape file resources used by plotly.py. These files are relatively large and change infrequently so it is useful to release them in a separate package.

Update version

Update the version of the plotly-geo package in packages/python/plotly-geo/setup.py.

This version is not intended to match the version of plotly.py.

Update CHANGELOG

Add a new entry to the CHANGELOG at packages/python/plotly-geo/CHANGELOG.md and commit the changes.

Tag Release

Create a new tag for the release

(plotly_dev) $ git checkout master
(plotly_dev) $ git stash
(plotly_dev) $ git pull origin master
(plotly_dev) $ git tag plotly-geo-vX.Y.Z
(plotly_dev) $ git push origin plotly-geo-vX.Y.Z

Publishing to PYPI

Publish the final version to PyPI

(plotly_dev) $ cd packages/python/plotly-geo
(plotly_dev) $ python setup.py sdist bdist_wheel
(plotly_dev) $ twine upload dist/plotly-geo-X.Y.Z.tar.gz
(plotly_dev) $ twine upload dist/plotly_geo-X.Y.Z-py3-none-any.whl

Publish to plotly anaconda channel

From packages/python/plotly-geo, build the conda package

(plotly_dev) $ conda build recipe/

Then upload to the plotly anaconda channel as described above

Release process - `chart-studio` package

The chart-studio package contains the utilities for interacting with Chart Studio (both Cloud or On-Prem).