Releasing a version#

The following guides include detailed information on how to prepare a NumPy release.

How to prepare a release#

These instructions give an overview of what is necessary to build binary releases for NumPy.

Current build and release info#

Useful info can be found in the following locations:

Supported platforms and versions#

NEP 29 outlines which Python versions are supported; For the first half of 2020, this will be Python >= 3.6. We test NumPy against all these versions every time we merge code to main. Binary installers may be available for a subset of these versions (see below).

  • OS X

    OS X versions >= 10.9 are supported, for Python version support see NEP 29. We build binary wheels for OSX that are compatible with Python, system Python, homebrew and macports - see this OSX wheel building summary for details.

  • Windows

    We build 32- and 64-bit wheels on Windows. Windows 7, 8 and 10 are supported. We build NumPy using the mingw-w64 toolchain, cibuildwheels and GitHub actions.

  • Linux

    We build and ship manylinux2014 wheels for NumPy. Many Linux distributions include their own binary builds of NumPy.

  • BSD / Solaris

    No binaries are provided, but successful builds on Solaris and BSD have been reported.

Tool chain#

We build all our wheels on cloud infrastructure - so this list of compilers is for information and debugging builds locally. See the .travis.yml script in the numpy wheels repo for an outdated source of the build recipes using multibuild.


The same gcc version is used as the one with which Python itself is built on each platform. At the moment this means:

  • OS X builds on travis currently use clang. It appears that binary wheels for OSX >= 10.6 can be safely built from the travis-ci OSX 10.9 VMs when building against the Python from the installers;

  • Windows builds use the mingw-w64 toolchain;

  • Manylinux2014 wheels use the gcc provided on the Manylinux docker images.

You will need Cython for building the binaries. Cython compiles the .pyx files in the NumPy distribution to .c files.


All the wheels link to a version of OpenBLAS supplied via the openblas-libs repo. The shared object (or DLL) is shipped with in the wheel, renamed to prevent name collisions with other OpenBLAS shared objects that may exist in the filesystem.

Building source archives and wheels#

The NumPy wheels and sdist are now built using cibuildwheel with github actions.

Building docs#

We are no longer building pdf files, only html docs. The needed to upload to the doc server can be built with spin docs dist.

To install the necessary doc build dependencies into your development environment, run pip install -r doc_requirements.txt.

Uploading to PyPI#

The only application needed for uploading is

  • twine (pip).

You will also need a PyPI token, which is best kept on a keyring. See the twine keyring documentation for how to do that.

Generating author/PR lists#

You will need a personal access token so that scripts can access the github NumPy repository.

  • gitpython (pip)

  • pygithub (pip)

What is released#

  • Wheels We currently support Python 3.8-3.10 on Windows, OSX, and Linux.

    • Windows: 32-bit and 64-bit wheels built using Github actions;

    • OSX: x64_86 and arm64 OSX wheels built using Github actions;

    • Linux: x64_86 and aarch64 Manylinux2014 wheels built using Github actions.

  • Other Release notes and changelog

  • Source distribution We build source releases in the .tar.gz format.

Release process#

Agree on a release schedule#

A typical release schedule is one beta, two release candidates and a final release. It’s best to discuss the timing on the mailing list first, in order for people to get their commits in on time, get doc wiki edits merged, etc. After a date is set, create a new maintenance/x.y.z branch, add new empty release notes for the next version in the main branch and update the Trac Milestones.

Make sure current branch builds a package correctly#

The CI builds wheels when a PR header begins with REL. Your last PR before releasing should be so marked and all the tests should pass. You can also do:

git clean -fxdq
python bdist_wheel
python sdist

For details of the build process itself, it is best to read the Step-by-Step Directions below.


The following steps are repeated for the beta(s), release candidates(s) and the final release.

Check deprecations#

Before the release branch is made, it should be checked that all deprecated code that should be removed is actually removed, and all new deprecations say in the docstring or deprecation warning what version the code will be removed.

Check the C API version number#

The C API version needs to be tracked in three places

  • numpy/core/

  • numpy/core/code_generators/cversions.txt

  • numpy/core/include/numpy/numpyconfig.h

There are three steps to the process.

  1. If the API has changed, increment the C_API_VERSION in The API is unchanged only if any code compiled against the current API will be backward compatible with the last released NumPy version. Any changes to C structures or additions to the public interface will make the new API not backward compatible.

  2. If the C_API_VERSION in the first step has changed, or if the hash of the API has changed, the cversions.txt file needs to be updated. To check the hash, run the script numpy/core/ and note the API hash that is printed. If that hash does not match the last hash in numpy/core/code_generators/cversions.txt the hash has changed. Using both the appropriate C_API_VERSION and hash, add a new entry to cversions.txt. If the API version was not changed, but the hash differs, you will need to comment out the previous entry for that API version. For instance, in NumPy 1.9 annotations were added, which changed the hash, but the API was the same as in 1.8. The hash serves as a check for API changes, but it is not definitive.

    If steps 1 and 2 are done correctly, compiling the release should not give a warning “API mismatch detect at the beginning of the build”.

  3. The numpy/core/include/numpy/numpyconfig.h will need a new NPY_X_Y_API_VERSION macro, where X and Y are the major and minor version numbers of the release. The value given to that macro only needs to be increased from the previous version if some of the functions or macros in the include files were deprecated.

The C ABI version number in numpy/core/ should only be updated for a major release.

Check the release notes#

Use towncrier to build the release note and commit the changes. This will remove all the fragments from doc/release/upcoming_changes and add doc/release/<version>-note.rst.

towncrier build –version “<version>” git commit -m”Create release note”

Check that the release notes are up-to-date.

Update the release notes with a Highlights section. Mention some of the following:

  • major new features

  • deprecated and removed features

  • supported Python versions

  • for SciPy, supported NumPy version(s)

  • outlook for the near future

Step-by-step directions#

This is a walkthrough of the NumPy 1.21.0 release on Linux, modified for building with GitHub Actions and cibuildwheels and uploading to the staging repository for NumPy. The commands can be copied into the command line, but be sure to replace 1.21.0 by the correct version. This should be read together with the general release guide.

Facility preparation#

Before beginning to make a release, use the *_requirements.txt files to ensure that you have the needed software. Most software can be installed with pip, but some will require apt-get, dnf, or whatever your system uses for software. You will also need a GitHub personal access token (PAT) to push the documentation. There are a few ways to streamline things:

  • Git can be set up to use a keyring to store your GitHub personal access token. Search online for the details.

  • You can use the keyring app to store the PyPI password for twine. See the online twine documentation for details.

Release preparation#

Add/drop Python versions#

When adding or dropping Python versions, three files need to be edited:

  • .github/workflows/wheels.yml # for github cibuildwheel

  • .travis.yml # for cibuildwheel aarch64 builds

  • # for classifier and minimum version check.

Make these changes in an ordinary PR against main and backport if necessary. Using the BLD: prefix (build label) for the commit summary will cause the wheel builds to be run so that the changes will be tested, We currently release wheels for new Python versions after the first Python rc once manylinux and cibuildwheel support it. For Python 3.11 we were able to release within a week of the rc1 announcement.

Backport Pull Requests#

Changes that have been marked for this release must be backported to the maintenance/1.21.x branch.

Update release documentation#

Four documents usually need to be updated or created before making a release:

  • The changelog

  • The release-notes

  • The .mailmap file

  • The doc/source/release.rst file

These changes should be made as an ordinary PR against the maintenance branch. After release all files except doc/source/release.rst will need to be forward ported to the main branch.

Generate the changelog#

The changelog is generated using the changelog tool:

$ python tools/ $GITHUB v1.20.0..maintenance/1.21.x > doc/changelog/1.21.0-changelog.rst

where GITHUB contains your GitHub access token. The text will need to be checked for non-standard contributor names and dependabot entries removed. It is also a good idea to remove any links that may be present in the PR titles as they don’t translate well to markdown, replace them with monospaced text. The non-standard contributor names should be fixed by updating the .mailmap file, which is a lot of work. It is best to make several trial runs before reaching this point and ping the malefactors using a GitHub issue to get the needed information.

Finish the release notes#

If this is the first release in a series the release note is generated, see the release note in doc/release/upcoming_changes/README.rst to see how to do this. Generating the release notes will also delete all the news fragment files in doc/release/upcoming_changes/.

The generated release note will always need some fixups, the introduction will need to be written, and significant changes should be called out. For patch releases the changelog text may also be appended, but not for the initial release as it is too long. Check previous release notes to see how this is done. Note that the :orphan: markup at the top, if present, will need changing to .. currentmodule:: numpy and the doc/source/release.rst index file will need updating.

Check the file#

Check that the file points to the correct release notes. It should have been updated after the last release, but if not, fix it now:

$ gvim

Release walkthrough#

Note that in the code snippets below, upstream refers to the root repository on GitHub and origin to its fork in your personal GitHub repositories. You may need to make adjustments if you have not forked the repository but simply cloned it locally. You can also edit .git/config and add upstream if it isn’t already present.

1. Prepare the release commit#

Checkout the branch for the release, make sure it is up to date, and clean the repository:

$ git checkout maintenance/1.21.x
$ git pull upstream maintenance/1.21.x
$ git submodule update
$ git clean -xdfq

Sanity check:

$ python3 -m spin test -m full

Tag the release and push the tag. This requires write permission for the numpy repository:

$ git tag -a -s v1.21.0 -m"NumPy 1.21.0 release"
$ git push upstream v1.21.0

If you need to delete the tag due to error:

$ git tag -d v1.21.0
$ git push --delete upstream v1.21.0

2. Build wheels#

Build wheels via cibuildwheel (preferred)#

Tagging the build at the beginning of this process will trigger a wheel build via cibuildwheel and upload wheels and an sdist to the staging repo. The CI run on github actions (for all x86-based and macOS arm64 wheels) takes about 1 1/4 hours. The CI run on travis (for aarch64) takes less time. You can check for uploaded files at the staging repository, but note that it is not closely synched with what you see of the running jobs.

If you wish to manually trigger a wheel build, you can do so:

  • On github actions -> Wheel builder there is a “Run workflow” button, click on it and choose the tag to build

  • On travis there is a “More Options” button, click on it and choose a branch to build. There does not appear to be an option to build a tag.

If a wheel build fails for unrelated reasons, you can rerun it individually:

  • On github actions select Wheel builder click on the commit that contains the build you want to rerun. On the left there is a list of wheel builds, select the one you want to rerun and on the resulting page hit the counterclockwise arrows button.

  • On travis select the failing build, which will take you to the travis job for that build. Hit the restart job button.

Note that if you do need to rerun jobs, you will need to delete the uploaded file, if any, in the anaconda staging repository, The old files will not be overwritten.

3. Download wheels#

When the wheels have all been successfully built and staged, download them from the Anaconda staging directory using the tools/ script:

$ cd ../numpy
$ mkdir -p release/installers
$ python3 tools/ 1.21.0

4. Generate the README files#

This needs to be done after all installers are downloaded, but before the pavement file is updated for continued development:

$ paver write_release

5. Reset the maintenance branch into a development state (skip for prereleases)#

Create release notes for next release and edit them to set the version. These notes will be a skeleton and have little content:

$ cp doc/source/release/template.rst doc/source/release/1.21.1-notes.rst
$ gvim doc/source/release/1.21.1-notes.rst
$ git add doc/source/release/1.21.1-notes.rst

Add new release notes to the documentation release list and update the RELEASE_NOTES variable in

$ gvim doc/source/release.rst

Commit the result:

$ git commit -a -m"REL: prepare 1.21.x for further development"
$ git push upstream HEAD

6. Upload to PyPI#

Upload to PyPI using twine. A recent version of twine of is needed after recent PyPI changes, version 3.4.1 was used here:

$ cd ../numpy
$ twine upload release/installers/*.whl
$ twine upload release/installers/numpy-1.21.0.tar.gz  # Upload last.

If one of the commands breaks in the middle, you may need to selectively upload the remaining files because PyPI does not allow the same file to be uploaded twice. The source file should be uploaded last to avoid synchronization problems that might occur if pip users access the files while this is in process, causing pip to build from source rather than downloading a binary wheel. PyPI only allows a single source distribution, here we have chosen the zip archive.

7. Upload files to github#

Go to, there should be a v1.21.0 tag, click on it and hit the edit button for that tag. There are two ways to add files, using an editable text window and as binary uploads. Start by editing the release/ that is translated from the rst version using pandoc. Things that will need fixing: PR lines from the changelog, if included, are wrapped and need unwrapping, links should be changed to monospaced text. Then copy the contents to the clipboard and paste them into the text window. It may take several tries to get it look right. Then

  • Upload release/installers/numpy-1.21.0.tar.gz as a binary file.

  • Upload release/README.rst as a binary file.

  • Upload doc/changelog/1.21.0-changelog.rst as a binary file.

  • Check the pre-release button if this is a pre-releases.

  • Hit the {Publish,Update} release button at the bottom.

8. Upload documents to (skip for prereleases)#


You will need a GitHub personal access token to push the update.

This step is only needed for final releases and can be skipped for pre-releases and most patch releases. make merge-doc clones the numpy/doc repo into doc/build/merge and updates it with the new documentation:

$ git clean -xdfq
$ git co v1.21.0
$ pushd doc
$ make docenv && source docenv/bin/activate
$ make merge-doc
$ pushd build/merge

If the release series is a new one, you will need to add a new section to the doc/build/merge/index.html front page just after the “insert here” comment:

$ gvim index.html +/'insert here'

Further, update the version-switcher json file to add the new release and update the version marked (stable):

$ gvim _static/versions.json

Otherwise, only the zip link should be updated with the new tag name. Since we are no longer generating pdf files, remove the line for the pdf files if present:

$ gvim index.html +/'tag v1.21'

You can “test run” the new documentation in a browser to make sure the links work:

$ firefox index.html  # or google-chrome, etc.

Update the stable link and update:

$ ln -sfn 1.21 stable
$ ls -l  # check the link

Once everything seems satisfactory, update, commit and upload the changes:

$ python3
$ git commit -a -m"Add documentation for v1.21.0"
$ git push
$ deactivate
$ popd
$ popd

9. Announce the release on (skip for prereleases)#

This assumes that you have forked

$ cd ../
$ git checkout main
$ git pull upstream main
$ git checkout -b announce-numpy-1.21.0
$ gvim content/en/
  • For all releases, go to the bottom of the page and add a one line link. Look to the previous links for example.

  • For the *.0 release in a cycle, add a new section at the top with a short description of the new features and point the news link to it.

commit and push:

$ git commit -a -m"announce the NumPy 1.21.0 release"
$ git push origin HEAD

Go to your Github fork and make a pull request.

10. Announce to mailing lists#

The release should be announced on the numpy-discussion, scipy-devel, scipy-user, and python-announce-list mailing lists. Look at previous announcements for the basic template. The contributor and PR lists are the same as generated for the release notes above. If you crosspost, make sure that python-announce-list is BCC so that replies will not be sent to that list.

11. Post-release tasks (skip for prereleases)#

Checkout main and forward port the documentation changes:

$ git checkout -b post-1.21.0-release-update
$ git checkout maintenance/1.21.x doc/source/release/1.21.0-notes.rst
$ git checkout maintenance/1.21.x doc/changelog/1.21.0-changelog.rst
$ git checkout maintenance/1.21.x .mailmap  # only if updated for release.
$ gvim doc/source/release.rst  # Add link to new notes
$ git status  # check status before commit
$ git commit -a -m"MAINT: Update main after 1.21.0 release."
$ git push origin HEAD

Go to GitHub and make a PR.

12. Update oldest-supported-numpy#

If this release is the first one to support a new Python version, or the first to provide wheels for a new platform or PyPy version, the version pinnings in should be updated. Either submit a PR with changes to setup.cfg there, or open an issue with info on needed changes.

Branch walkthrough#

This guide contains a walkthrough of branching NumPy 1.21.x on Linux. The commands can be copied into the command line, but be sure to replace 1.21 and 1.22 by the correct versions. It is good practice to make .mailmap as current as possible before making the branch, that may take several weeks.

This should be read together with the general release guide.


Make the branch#

This is only needed when starting a new maintenance branch. Because NumPy now depends on tags to determine the version, the start of a new development cycle in the main branch needs an annotated tag. That is done as follows:

$ git checkout main
$ git pull upstream main
$ git commit --allow-empty -m'REL: Begin NumPy 1.22.0 development'
$ git push upstream HEAD

If the push fails because new PRs have been merged, do:

$ git pull --rebase upstream

and repeat the push. Once the push succeeds, tag it:

$ git tag -a -s v1.22.0.dev0 -m'Begin NumPy 1.22.0 development'
$ git push upstream v1.22.0.dev0

then make the new branch and push it:

$ git branch maintenance/1.21.x HEAD^
$ git push upstream maintenance/1.21.x

Prepare the main branch for further development#

Make a PR branch to prepare main for further development:

$ git checkout -b 'prepare-main-for-1.22.0-development' v1.22.0.dev0

Delete the release note fragments:

$ git rm doc/release/upcoming_changes/[0-9]*.*.rst

Create the new release notes skeleton and add to index:

$ cp doc/source/release/template.rst doc/source/release/1.22.0-notes.rst
$ gvim doc/source/release/1.22.0-notes.rst  # put the correct version
$ git add doc/source/release/1.22.0-notes.rst
$ gvim doc/source/release.rst  # add new notes to notes index
$ git add doc/source/release.rst

Update and update the RELEASE_NOTES variable to point to the new notes:

$ gvim
$ git add

Update cversions.txt to add current release. There should be no new hash to worry about at this early point, just add a comment following previous practice:

$ gvim numpy/core/code_generators/cversions.txt
$ git add numpy/core/code_generators/cversions.txt

Check your work, commit it, and push:

$ git status  # check work
$ git commit -m'REL: Prepare main for NumPy 1.22.0 development'
$ git push origin HEAD

Now make a pull request.