cibuildwheel

Documentation

Python wheels are great. Building them across Mac, Linux, Windows, on multiple versions of Python, is not.

cibuildwheel is here to help. cibuildwheel runs on your CI server - currently it supports GitHub Actions, Azure Pipelines, Travis CI, AppVeyor, CircleCI, and GitLab CI - and it builds and tests your wheels across all of your platforms.

What does it do?

^{¹ PyPy is only supported for manylinux wheels.
² Windows arm64 support is experimental.
³ Alpine 3.14 and very briefly 3.15's default python3 was not able to load musllinux wheels. This has been fixed; please upgrade the python package if using Alpine from before the fix.
⁴ Cross-compilation not supported with PyPy - to build these wheels you need to run cibuildwheel on an Apple Silicon machine.
⁵ CPython 3.12 is built by default using Python RCs, starting with cibuildwheel 2.15.}

• Builds manylinux, musllinux, macOS 10.9+, and Windows wheels for CPython and PyPy
• Works on GitHub Actions, Azure Pipelines, Travis CI, AppVeyor, CircleCI, GitLab CI, and Cirrus CI
• Bundles shared library dependencies on Linux and macOS through auditwheel and delocate
• Runs your library's tests against the wheel-installed version of your library

See the cibuildwheel 1 documentation if you need to build unsupported versions of Python, such as Python 2.

Usage

cibuildwheel runs inside a CI service. Supported platforms depend on which service you're using:

^{¹ Requires emulation, distributed separately. Other services may also support Linux ARM through emulation or third-party build hosts, but these are not tested in our CI.
² Uses cross-compilation. It is not possible to test arm64 and the arm64 part of a universal2 wheel on this CI platform.
³ Uses cross-compilation. Thanks to Rosetta 2 emulation, it is possible to test x86_64 and both parts of a universal2 wheel on this CI platform.
⁴ Uses cross-compilation. It is not possible to test arm64 on this CI platform.}

Example setup

To build manylinux, musllinux, macOS, and Windows wheels on GitHub Actions, you could use this .github/workflows/wheels.yml:

name: Build

on: [push, pull_request]

jobs:
  build_wheels:
    name: Build wheels on ${{ matrix.os }}
    runs-on: ${{ matrix.os }}
    strategy:
      matrix:
        os: [ubuntu-20.04, windows-2019, macOS-11]

    steps:
      - uses: actions/checkout@v4

      # Used to host cibuildwheel
      - uses: actions/setup-python@v3

      - name: Install cibuildwheel
        run: python -m pip install cibuildwheel==2.16.2

      - name: Build wheels
        run: python -m cibuildwheel --output-dir wheelhouse
        # to supply options, put them in 'env', like:
        # env:
        #   CIBW_SOME_OPTION: value

      - uses: actions/upload-artifact@v3
        with:
          path: ./wheelhouse/*.whl

For more information, including PyPI deployment, and the use of other CI services or the dedicated GitHub Action, check out the documentation and the examples.

How it works

The following diagram summarises the steps that cibuildwheel takes on each platform.

^{Explore an interactive version of this diagram in the docs.}

Options

These options can be specified in a pyproject.toml file, as well; see configuration.

Working examples

Here are some repos that use cibuildwheel.

Name	CI	OS	Notes
scikit-learn			The machine learning library. A complex but clean config using many of cibuildwheel's features to build a large project with Cython and C++ extensions.
pytorch-fairseq			Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
NumPy			The fundamental package for scientific computing with Python.
Tornado			Tornado is a Python web framework and asynchronous networking library. Uses stable ABI for a small C extension.
Matplotlib			The venerable Matplotlib, a Python library with C++ portions
NCNN			ncnn is a high-performance neural network inference framework optimized for the mobile platform
Prophet			Tool for producing high quality forecasts for time series data that has multiple seasonality with linear or non-linear growth.
MyPy			The compiled version of MyPy using MyPyC.
Kivy			Open source UI framework written in Python, running on Windows, Linux, macOS, Android and iOS
duckdb			DuckDB is an in-process SQL OLAP Database Management System

ℹ️ That's just a handful, there are many more! Check out the Working Examples page in the docs.

Legal note

Since cibuildwheel repairs the wheel with delocate or auditwheel, it might automatically bundle dynamically linked libraries from the build machine.

It helps ensure that the library can run without any dependencies outside of the pip toolchain.

This is similar to static linking, so it might have some license implications. Check the license for any code you're pulling in to make sure that's allowed.

Changelog

v2.16.2

3 October 2023

• 🛠 Updates CPython 3.12 version to 3.12.0, final release (#1635)
• ✨ Adds a debug option CIBW_DEBUG_KEEP_CONTAINER to stop cibuildwheel deleting build containers after the build finishes. (#1620)
• 📚 Adds support for [tool.cibuildwheel] checking by adding a schema compatible with the validate-pyproject tool (#1622, #1628, #1629)
• 🐛 Fix parsing of CIBW_CONTAINER_ENGINE and CIBW_BUILD_FRONTEND options to not break arguments on : characters (#1621)
• 🐛 Fix the evaluation order of CIBW_ENVIRONMENT and CIBW_ENVIRONMENT_PASS so that CIBW_ENVIRONMENT assignments can reference environment variables passed through from the host machine. (#1617)
• 🛠 Supports manylinux images' deferred installation of interpreters through the manylinux-interpreters tool (#1630)

v2.16.1

26 September 2023

• 🛠 Updates the prerelease CPython 3.12 version to 3.12.0rc3 (#1625)
• 🛠 Only calls linux32 in containers when necessary (#1599)

v2.16.0

18 September 2023

• ✨ Add the ability to pass additional flags to a build frontend through the CIBW_BUILD_FRONTEND option (#1588).
• ✨ The environment variable SOURCE_DATE_EPOCH is now automatically passed through to container Linux builds (useful for reproducible builds!) (#1589)
• 🛠 Updates the prerelease CPython 3.12 version to 3.12.0rc2 (#1604)
• 🐛 Fix requires_python auto-detection from setup.py when the call to setup() is within an if __name__ == "__main__" block (#1613)
• 🐛 Fix a bug that prevented building Linux wheels in Docker on a Windows host (#1573)
• 🐛 --only can now select prerelease-pythons (#1564)
• 📚 Docs & examples updates (#1582, #1593, #1598, #1615)

v2.15.0

8 August 2023

• 🌟 CPython 3.12 wheels are now built by default - without the CIBW_PRERELEASE_PYTHONS flag. It's time to build and upload these wheels to PyPI! This release includes CPython 3.12.0rc1, which is guaranteed to be ABI compatible with the final release. (#1565)
• ✨ Adds musllinux_1_2 support - this allows packagers to build for musl-based Linux distributions on a more recent Alpine image, and a newer musl libc. (#1561)

v2.14.1

15 July 2023

• 🛠 Updates the prerelease CPython 3.12 version to 3.12.0b4 (#1550)

That's the last few versions.

ℹ️ Want more changelog? Head over to the changelog page in the docs.

Contributing

For more info on how to contribute to cibuildwheel, see the docs.

Everyone interacting with the cibuildwheel project via codebase, issue tracker, chat rooms, or otherwise is expected to follow the PSF Code of Conduct.

Maintainers

• Joe Rickerby @joerick
• Yannick Jadoul @YannickJadoul
• Matthieu Darbois @mayeut
• Henry Schreiner @henryiii
• Grzegorz Bokota @Czaki

Credits

cibuildwheel stands on the shoulders of giants.

• ⭐️ @matthew-brett for multibuild and matthew-brett/delocate
• @PyPA for the manylinux Docker images pypa/manylinux
• @ogrisel for wheelhouse-uploader and run_with_env.cmd

Massive props also to-

• @zfrenchee for help debugging many issues
• @lelit for some great bug reports and contributions
• @mayeut for a phenomenal PR patching Python itself for better compatibility!
• @czaki for being a super-contributor over many PRs and helping out with countless issues!
• @mattip for his help with adding PyPy support to cibuildwheel

See also

Another very similar tool to consider is matthew-brett/multibuild. multibuild is a shell script toolbox for building a wheel on various platforms. It is used as a basis to build some of the big data science tools, like SciPy.

If you are building Rust wheels, you can get by without some of the tricks required to make GLIBC work via manylinux; this is especially relevant for cross-compiling, which is easy with Rust. See maturin-action for a tool that is optimized for building Rust wheels and cross-compiling.