Run cibuildwheel locally (optional)

Before getting to CI setup, it can be convenient to test cibuildwheel locally to quickly iterate and track down issues. If you've got Docker installed on your development machine, you can run a Linux build without even touching CI.

Tip

You can run the Linux build on any platform. Even Windows can run Linux containers these days, but there are a few hoops to jump through. Check this document for more info.

This is convenient as it's quicker to iterate, and because the builds are happening in manylinux Docker containers, they're perfectly reproducible.

Install cibuildwheel and run a build like this:

pip install cibuildwheel
cibuildwheel --platform linux

Or, using pipx:

pipx run cibuildwheel --platform linux

You should see the builds taking place. You can experiment with options using environment variables or pyproject.toml.

Environment variables

cibuildwheel will read config from the environment. Syntax varies, depending on your shell:

POSIX shell (Linux/macOS)

# run a command to set up the build system
export CIBW_BEFORE_ALL='apt install libpng-dev'

cibuildwheel --platform linux

CMD (Windows)

set CIBW_BEFORE_ALL='apt install libpng-dev'

cibuildwheel --platform linux

pyproject.toml

If you write your options into pyproject.toml, you can work on your options locally, and they'll be automatically picked up when running in CI.

pyproject.toml

[tool.cibuildwheel]
before-all = "apt install libpng-dev"

Then invoke cibuildwheel, like:

cibuildwheel --platform linux

Configure a CI service

GitHub Actions [linux/mac/windows]

To build Linux, Mac, and Windows wheels using GitHub Actions, create a .github/workflows/build_wheels.yml file in your repo.

Action

For GitHub Actions, cibuildwheel provides an action you can use. This is concise and enables easier auto updating via GitHub's Dependabot; see Automatic updates.

.github/workflows/build_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-10.15]

    steps:
      - uses: actions/checkout@v2

      - name: Build wheels
        uses: pypa/cibuildwheel@v2.1.2
        # to supply options, put them in 'env', like:
        # env:
        #   CIBW_SOME_OPTION: value

      - uses: actions/upload-artifact@v2
        with:
          path: ./wheelhouse/*.whl

You can use env: with the action just like you would with run:; you can also use with: to set the command line options: package-dir: . and output-dir: wheelhouse (those values are the defaults).

pipx

The GitHub Actions runners have pipx installed, so you can easily build in just one line. This is internally how the action works; the main benefit of the action form is easy updates via GitHub's Dependabot.

.github/workflows/build_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-10.15]

    steps:
      - uses: actions/checkout@v2

      - name: Build wheels
        run: pipx run cibuildwheel==2.1.2

      - uses: actions/upload-artifact@v2
        with:
          path: ./wheelhouse/*.whl

Generic

This is the most generic form using setup-python and pip; it looks the most like the other CI examples. If you want to avoid having setup that takes advantage of GitHub Actions features or pipx being preinstalled, this might appeal to you.

.github/workflows/build_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-10.15]

    steps:
      - uses: actions/checkout@v2

      # Used to host cibuildwheel
      - uses: actions/setup-python@v2

      - name: Install cibuildwheel
        run: python -m pip install cibuildwheel==2.1.2

      - name: Build wheels
        run: python -m cibuildwheel --output-dir wheelhouse

      - uses: actions/upload-artifact@v2
        with:
          path: ./wheelhouse/*.whl

Commit this file, and push to GitHub - either to your default branch, or to a PR branch. The build should start automatically.

For more info on this file, check out the docs.

examples/github-deploy.yml extends this minimal example with a demonstration of how to automatically upload the built wheels to PyPI.

Azure Pipelines [linux/mac/windows]

To build Linux, Mac, and Windows wheels on Azure Pipelines, create a azure-pipelines.yml file in your repo.

azure-pipelines.yml

jobs:
- job: linux
  pool: {vmImage: 'Ubuntu-16.04'}
  steps:
    - task: UsePythonVersion@0
    - bash: |
        set -o errexit
        python3 -m pip install --upgrade pip
        pip3 install cibuildwheel==2.1.2
      displayName: Install dependencies
    - bash: cibuildwheel --output-dir wheelhouse .
      displayName: Build wheels
    - task: PublishBuildArtifacts@1
      inputs: {pathtoPublish: 'wheelhouse'}

- job: macos
  pool: {vmImage: 'macOS-10.15'}
  steps:
    - task: UsePythonVersion@0
    - bash: |
        set -o errexit
        python3 -m pip install --upgrade pip
        python3 -m pip install cibuildwheel==2.1.2
      displayName: Install dependencies
    - bash: cibuildwheel --output-dir wheelhouse .
      displayName: Build wheels
    - task: PublishBuildArtifacts@1
      inputs: {pathtoPublish: wheelhouse}

- job: windows
  pool: {vmImage: 'vs2017-win2016'}
  steps:
    - task: UsePythonVersion@0
    - bash: |
        set -o errexit
        python -m pip install --upgrade pip
        pip install cibuildwheel==2.1.2
      displayName: Install dependencies
    - bash: cibuildwheel --output-dir wheelhouse .
      displayName: Build wheels
    - task: PublishBuildArtifacts@1
      inputs: {pathtoPublish: 'wheelhouse'}

Commit this file, enable building of your repo on Azure Pipelines, and push.

Wheels will be stored for you and available through the Pipelines interface. For more info on this file, check out the docs.

Travis CI [linux/windows]

To build Linux and Windows wheels on Travis CI, create a .travis.yml file in your repo.

.travis.yml

os: linux
dist: focal
language: python

jobs:
  include:
    # perform a linux build
    - services: docker
    # perform a linux ARMv8 build
    - services: docker
      arch: arm64
    # perform a linux PPC64LE build
    - services: docker
      arch: ppc64le
    # perform a linux S390X build
    - services: docker
      arch: s390x
    # and a windows build
    - os: windows
      language: shell
      before_install:
        - choco upgrade python -y --version 3.8.6
        - export PATH="/c/Python38:/c/Python38/Scripts:$PATH"
        # make sure it's on PATH as 'python3'
        - ln -s /c/Python38/python.exe /c/Python38/python3.exe

install:
  - python3 -m pip install cibuildwheel==2.1.2

script:
  # build the wheels, put them into './wheelhouse'
  - python3 -m cibuildwheel --output-dir wheelhouse

Commit this file, enable building of your repo on Travis CI, and push.

Then setup a deployment method by following the Travis CI deployment docs, or see Delivering to PyPI. For more info on .travis.yml, check out the docs.

examples/travis-ci-deploy.yml extends this minimal example with a demonstration of how to automatically upload the built wheels to PyPI.

AppVeyor [linux/mac/windows]

To build Linux, Mac, and Windows wheels on AppVeyor, create an appveyor.yml file in your repo.

appveyor.yml

environment:
  matrix:
    - APPVEYOR_BUILD_WORKER_IMAGE: Ubuntu
      APPVEYOR_JOB_NAME: "python37-x64-ubuntu"
    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
      APPVEYOR_JOB_NAME: "python37-x64-vs2015"
    - APPVEYOR_BUILD_WORKER_IMAGE: macos-mojave
      APPVEYOR_JOB_NAME: "python37-x64-macos-mojave"

stack: python 3.7

init:
- cmd: set PATH=C:\Python37;C:\Python37\Scripts;%PATH%

install: python -m pip install cibuildwheel==2.1.2

build_script: python -m cibuildwheel --output-dir wheelhouse

artifacts:
  - path: "wheelhouse\\*.whl"
    name: Wheels

Commit this file, enable building of your repo on AppVeyor, and push.

AppVeyor will store the built wheels for you - you can access them from the project console. Alternatively, you may want to store them in the same place as the Travis CI build. See AppVeyor deployment docs for more info, or see Delivering to PyPI below.

For more info on this config file, check out the docs.

CircleCI [linux/mac]

To build Linux and Mac wheels on CircleCI, create a .circleci/config.yml file in your repo,

.circleci/config.yml

version: 2

jobs:
  linux-wheels:
    working_directory: ~/linux-wheels
    docker:
      - image: circleci/python:3.6
    steps:
      - checkout
      - setup_remote_docker
      - run:
          name: Build the Linux wheels.
          command: |
            pip3 install --user cibuildwheel==2.1.2
            cibuildwheel --output-dir wheelhouse
      - store_artifacts:
          path: wheelhouse/

  osx-wheels:
    working_directory: ~/osx-wheels
    macos:
      xcode: 12.5.1
    steps:
      - checkout
      - run:
          name: Build the OS X wheels.
          command: |
            pip3 install cibuildwheel==2.1.2
            cibuildwheel --output-dir wheelhouse
      - store_artifacts:
          path: wheelhouse/

workflows:
  version: 2
  all-tests:
    jobs:
      - linux-wheels
      - osx-wheels

Commit this file, enable building of your repo on CircleCI, and push.

Note

CircleCI doesn't enable free macOS containers for open source by default, but you can ask for access. See here for more information.

CircleCI will store the built wheels for you - you can access them from the project console. Check out the CircleCI docs for more info on this config file.

Gitlab CI [linux]

To build Linux wheels on Gitlab CI, create a .gitlab-ci.yml file in your repo,

.gitlab-ci.yml

linux:
  image: python:3.8
  # make a docker daemon available for cibuildwheel to use
  services:
    - name: docker:dind
      entrypoint: ["env", "-u", "DOCKER_HOST"]
      command: ["dockerd-entrypoint.sh"]
  variables:
    DOCKER_HOST: tcp://docker:2375/
    DOCKER_DRIVER: overlay2
    # See https://github.com/docker-library/docker/pull/166
    DOCKER_TLS_CERTDIR: ""
  script:
    - curl -sSL https://get.docker.com/ | sh
    - python -m pip install cibuildwheel==2.1.2
    - cibuildwheel --output-dir wheelhouse
  artifacts:
    paths:
      - wheelhouse/

Commit this file, and push to Gitlab. The pipeline should start automatically.

Gitlab will store the built wheels for you - you can access them from the Pipelines view. Check out the Gitlab docs for more info on this config file.

⚠️ Got an error? Check the FAQ.

Next steps

Once you've got the wheel building successfully, you might want to set up testing or automatic releases to PyPI.