Release guide#

This guide documents napari’s release process. Currently, it only handles distribution, but as the project matures, it will include generating release notes, documentation, etc.


New versions of napari will be released every two months. The first release candidate will be available one week prior to release for testing purposes. Multiple release candidates may become available during the week prior to release. Upcoming releases can be found in our public calendar.

The latest release candidate can be installed with

python -m pip install --pre napari

Release management#

The release will be coordinated by a release manager whose responsibilities include…

Two weeks before release (one week before release candidate)#

  • Look through currently open PRs and get a sense of what would be good to merge before the first release candidate

  • Ensure conda-recipe/meta.yaml in napari/packaging is up-to-date (e.g. run dependencies match setup.cfg requirements).

  • Create a zulip thread in the release channel letting people know the release candidate is coming and pointing out PRs that would be nice to merge before release

At this stage, bug fixes and features that are close to landing should be prioritized. The release manager will follow up with PR authors, reviewing and merging as needed.

Nine days before release (two days before release candidate)#

  • Generate release notes with the script in the release folder

  • Fill in the release highlights and make a PR with the release notes

At this point the release manager should ideally be the only person merging PRs on the repo for the next week.

One week before release#

  • Add any recently merged PRs to release notes

  • Merge release notes

  • Make the release candidate

  • Announce to release stream on zulip that the first release candidate is available for testing

The week before release#

  • Merge any PRs and update release notes accordingly

  • Make new release candidates as necessary and announce them on zulip

At this stage PRs merged should focus mainly on regressions and bug fixes. New features should wait until after release.

The day of release#

  • make sure final rc has been tested

  • ensure all PRs have been added to release notes and then make release and announce on zulip

Release process#

Additional release dependencies (python -m pip install -e .[release]) are required to complete the release process. determines which non-Python files are included. Make sure to check that all necessary ones are listed before beginning the release process.

The napari/napari repository must have a PyPI API token as a GitHub secret. This likely has been done already, but if it has not, follow this guide to gain a token and this guide to add it as a secret.

Determining the version#

The version of napari is automatically determined at install time by setuptools_scm from the latest git tag beginning with v. Thus, you’ll need to tag the reference with the new version number. It is likely something like X.Y.Z. Before making a release though we need to generate the release notes.

Generating release notes#

  1. Make a list of merges, contributors, and reviewers by running python docs/release/ -h and following that file’s usage. For each release generate the list to include everything since the last release for which there are release notes (which should just be the last release). For example making the release notes for the 0.2.1 release can be done as follows:

    python docs/release/ v0.2.0 main --version 0.2.1 | tee docs/release/
  2. Scan the PR titles for highlights, deprecations, API changes, and bugfixes, and mention these in the relevant sections of the notes. Try to present the information in an expressive way by mentioning the affected functions, elaborating on the changes and their consequences. If possible, organize semantically close PRs in groups.

  3. Make sure the file name is of the form doc/release/release_<major>_<minor>_<release>.md.

  4. Make and merge a PR with these release notes before moving onto the next steps.

Update translation strings#

As new code is included in the codebase, some of the strings that need to be translated might not yet be using the trans methods. To help keep the codebase up to date in terms of translations we added a test script that runs daily on CI and can be also run locally to ensure that a release includes the most up to date translatable strings.

The test script is available on the /tools/ file and it relies on an additional file /tools/ to include strings to skip safely from translation.

The test checks:

  1. Untranslated strings: not using the trans methods.

  2. Outdated skip strings: should no longer be included in the /tools/ file.

  3. Translation usage errors: where translation strings may be missing interpolation variables.

You can execute tests locally from the repository root, and follow the instructions printed on the stdout if any test fails.

pytest tools/ --tb=short

Tagging the new release candidate#

First we will generate a release candidate, which will contain the letters rc. Using release candidates allows us to test releases on PyPI without using up the actual release number.

You can tag the current source code as a release candidate with:

git tag vX.Y.Zrc1 main

If the tag is meant for a previous version of main, simply reference the specific commit:

git tag vX.Y.Zrc1 abcde42

Note here how we are using rc for release candidate to create a version of our release we can test before making the real release.

You can read more on tagging here.

Testing the release candidate#

Our CI automatically makes a release, copying the release notes to the tag and uploading the distribution to PyPI. You can trigger this by pushing the new tag to napari/napari:

git push upstream --tags

The release candidate can then be tested with

python -m pip install --pre napari

It is recommended that the release candidate is tested in a virtual environment in order to isolate dependencies.

If the release candidate is not what you want, make your changes and repeat the process from the beginning but incrementing the number after rc on tag (e.g. vX.Y.Zrc2).

Once you are satisfied with the release candidate it is time to generate the actual release.

Generating the actual release#

To generate the actual release you will now repeat the processes above but now dropping the rc. For example:

git tag vX.Y.Z main
git push upstream --tags

conda-forge packages#

The packages on conda-forge are not controlled directly by our repositories. Instead, they are governed by the conda-forge/napari-feedstock repository. The essential actions are automated, but there are a few maintenance notes we need to have in mind.

New releases#

Once the PyPI release is available, the conda-forge bots will submit a PR to conda-forge/napari-feedstock within a few hours. Merging that PR to main will trigger the conda-forge release. Accounting for the build times and the CDN sync, this means that the conda-forge packages will be available 30-60 mins after the PR is merged.

Before merging, please pay special attention to these aspects:

  • Version string has been correctly updated. The build number should have been reset to 0 now.

  • The CI passes correctly. Do check the logs, especially the test section (search for TEST START).

  • The run dependencies match the runtime requirements of the PyPI release (listed in setup.cfg). Watch for modified version constraints, as well as added or removed packages. Note that the conda-forge packages include some more dependencies for convenience, so you might need to check the extras sections in setup.cfg.


See these PRs for examples on previous conda-forge releases:

Patch dependencies of previous releases#

conda-forge offers a mechanism to patch the metadata of existing releases. This is useful when a new dependency release breaks napari in some way or, in general, when the metadata of an existing package is proven wrong after it has been released.

To amend the metadata, we need to:

Some previous examples include:

Broken packages#

In some cases, a wrongly merged PR might cause the release of a broken artifact. If this is not fixable with a metadata patch (see above), then the packages can be marked as broken. To do so, we can submit a PR to conda-forge/admin-requests.

For more details, follow the instructions for “Mark packages as broken on conda-forge”.

Please make sure a correct build for the problematic release is available before (or shortly after) the admin-requests PR is merged!