Tips

Linux builds on Docker

Linux wheels are built in the manylinux docker images to provide binary compatible wheels on Linux, according to PEP 571. Because of this, when building with cibuildwheel on Linux, a few things should be taken into account:

  • Programs and libraries are not installed on the CI runner host, but rather should be installed inside of the Docker image - using yum for manylinux2010 or manylinux2014, and apt-get for manylinux_2_24, or manually. The same goes for environment variables that are potentially needed to customize the wheel building.

    cibuildwheel supports this by providing the CIBW_ENVIRONMENT and CIBW_BEFORE_ALL options to setup the build environment inside the running Docker image.

  • The project directory is mounted in the running Docker instance as /project, the output directory for the wheels as /output. In general, this is handled transparently by cibuildwheel. For a more finegrained level of control however, the root of the host file system is mounted as /host, allowing for example to access shared files, caches, etc. on the host file system. Note that /host is not available on CircleCI due to their Docker policies.

  • Alternative Docker images can be specified with the CIBW_MANYLINUX_*_IMAGE options to allow for a custom, preconfigured build environment for the Linux builds. See options for more details.

Building macOS wheels for Apple Silicon

cibuildwheel supports cross-compiling universal2 and arm64 wheels on x86_64 runners. With the introduction of Apple Silicon, you now have several choices for wheels for Python 3.8+:

x86_64

The traditional wheel for Apple, loads on Intel machines, and on Apple Silicon when running Python under Rosetta 2 emulation.

Due to a change in naming, Pip 20.3+ (or an installer using packaging 20.5+) is required to install a binary wheel on macOS Big Sur.

arm64

The native wheel for macOS on Apple Silicon.

Requires Pip 20.3+ (or packaging 20.5+) to install.

universal2

This wheel contains both architectures, causing it to be up to twice the size (data files do not get doubled, only compiled code). It requires Pip 20.3 (Packaging 20.6+) to load on Intel, and Pip 21.0.1 (Packaging 20.9+) to load on Apple Silicon.

Note

The dual-architecture universal2 has a few benefits, but a key benefit to a universal wheel is that a user can bundle these wheels into an application and ship a single binary.

However, if you have a large library, then you might prefer to ship the two single-arch wheels instead - x86_64 and arm64. In rare cases, you might want to build all three, but in that case, pip will not download the universal wheels, because it prefers the most specific wheel available.

Generally speaking, because Pip 20.3 is required for the universal2 wheel, most packages should provide both x86_64 and universal2 wheels for now. Once Pip 20.3+ is common on macOS, then it should be possible to ship only the universal2 wheel.

Apple Silicon wheels are not built by default, but can be enabled by adding extra archs to the CIBW_ARCHS_MACOS option - e.g. x86_64 arm64 universal2. Cross-compilation is provided by the Xcode toolchain.

Important

When cross-compiling on Intel, it is not possible to test arm64 and the arm64 part of a universal2 wheel.

cibuildwheel will raise a warning to notify you of this - these warnings be be silenced by skipping testing on these platforms: CIBW_TEST_SKIP: *_arm64 *_universal2:arm64.

Hopefully, cross-compilation is a temporary situation. Once we have widely available Apple Silicon CI runners, we can build and test arm64 and universal2 wheels natively. That's why universal2 wheels are not yet built by default, and require opt-in by setting CIBW_ARCHS_MACOS.

Note

Your runner needs Xcode Command Line Tools 12.2 or later to build universal2 or arm64.

Only CPython 3.8 and newer support universal2 and arm64 wheels.

Here's an example GitHub Actions workflow with a job that builds for Apple Silicon:

.github/workflows/build_macos.yml

name: Build

on: [push, pull_request]

jobs:
  build_wheels_macos:
    name: Build wheels on macos-10.15
    runs-on: macos-10.15
    steps:
      - uses: actions/checkout@v2

      - name: Build wheels
        uses: pypa/cibuildwheel@v2.1.2
        env:
          CIBW_ARCHS_MACOS: x86_64 universal2

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

Building non-native architectures using emulation

cibuildwheel supports building non-native architectures on Linux, via emulation through the binfmt_misc kernel feature. The easiest way to use this is via the docker/setup-qemu-action on GitHub Actions or tonistiigi/binfmt.

Check out the following config for an example of how to set it up on GitHub Actions. Once QEMU is set up and registered, you just need to set the CIBW_ARCHS_LINUX environment variable (or use the --archs option on Linux), and the other architectures are emulated automatically.

.github/workflows/build.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

      - uses: actions/setup-python@v2
        name: Install Python
        with:
          python-version: '3.7'

      - name: Set up QEMU
        if: runner.os == 'Linux'
        uses: docker/setup-qemu-action@v1
        with:
          platforms: all

      - name: Build wheels
        uses: pypa/cibuildwheel@v2.1.2
        env:
          # configure cibuildwheel to build native archs ('auto'), and some
          # emulated ones
          CIBW_ARCHS_LINUX: auto aarch64 ppc64le s390x

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

Building packages with optional C extensions

cibuildwheel defines the environment variable CIBUILDWHEEL to the value 1 allowing projects for which the C extension is optional to make it mandatory when building wheels.

An easy way to do it in Python 3 is through the optional named argument of Extension constructor in your setup.py:

myextension = Extension(
    "myextension",
    ["myextension.c"],
    optional=os.environ.get('CIBUILDWHEEL', '0') != '1',
)

Automatic updates

Selecting a moving target (like the latest release) is generally a bad idea in CI. If something breaks, you can't tell whether it was your code or an upstream update that caused the breakage, and in a worse-case scenario, it could occur during a release. There are two suggested methods for keeping cibuildwheel up to date that instead involve scheduled pull requests using GitHub's dependabot.

Option 1: GitHub Action

If you use GitHub Actions for builds, you can use cibuildwheel as an action:

uses: pypa/cibuildwheel@v2.1.2

This is a composite step that just runs cibuildwheel using pipx. You can set command-line options as with: parameters, and use env: as normal.

Then, your .github/dependabot.yml file could look like this:

version: 2
updates:
  - package-ecosystem: "github-actions"
    directory: "/"
    schedule:
      interval: "weekly"
    ignore:
      # Optional: Official actions have moving tags like v1;
      # if you use those, you don't need updates.
      - dependency-name: "actions/*"

Option 2: Requirement files

The second option, and the only one that supports other CI systems, is using a requirements-*.txt file. The file should have a distinct name and have only one entry:

# requirements-cibw.txt
cibuildwheel==2.1.2

Then your install step would have python -m pip install -r requirements-cibw.txt in it. Your .github/dependabot.yml file could look like this:

version: 2
updates:
  - package-ecosystem: "pip"
    directory: "/"
    schedule:
      interval: "daily"

This will also try to update other pins in all requirement files, so be sure you want to do that. The only control you have over the files used is via the directory option.

Alternatives to cibuildwheel options

cibuildwheel provides lots of opportunities to configure the build environment. However, you might consider adding this build configuration into the package itself - in general, this is preferred, because users of your package 'sdist' will also benefit.

Missing build dependencies

If your build needs Python dependencies, rather than using CIBW_BEFORE_BUILD, it's best to add these to the build-system.requires section of your pyproject.toml. For example, if your project requires Cython to build, your pyproject.toml might include a section like this:

[build-system]
requires = [
    "setuptools>=42",
    "wheel",
    "Cython",
]

build-backend = "setuptools.build_meta"

Actions you need to perform before building

You might need to run some other commands before building, like running a script that performs codegen or downloading some data that's not stored in your source tree.

Rather than using CIBW_BEFORE_ALL or CIBW_BEFORE_BUILD, you could incorporate these steps into your package's build process. For example, if you're using setuptools, you can add steps to your package's setup.py using a structure like this:

```python
import subprocess
import setuptools
import setuptools.command.build_py


class BuildPyCommand(setuptools.command.build_py.build_py):
  """Custom build command."""

  def run(self):
    # your custom build steps here
    # e.g.
    #   subprocess.run(['python', 'scripts/my_custom_script.py'], check=True)
    setuptools.command.build_py.build_py.run(self)


setuptools.setup(
    cmdclass={
        'build_py': BuildPyCommand,
    },
    # Usual setup() args.
    # ...
)
```

Compiler flags

Your build might need some compiler flags to be set through environment variables. Consider incorporating these into your package, for example, in setup.py using extra_compile_args or extra_link_args.

Python 2.7 / PyPy2 wheels

See the cibuildwheel version 1 docs for information about building Python 2.7 or PyPy2 wheels. There are lots of tricks and workaround there that are no longer required for Python 3 in cibuildwheel 2.

Troubleshooting

If your wheel didn't compile, you might have a mistake in your config.

To quickly test your config without doing a git push and waiting for your code to build on CI, you can test the Linux build in a local Docker container.

Missing dependencies

You might need to install something on the build machine. You can do this with apt/yum, brew or choco, using the CIBW_BEFORE_ALL option. Or, for a Python dependency, consider adding it to pyproject.toml.

macOS: ModuleNotFoundError

Calling cibuildwheel from a python3 script and getting a ModuleNotFoundError? Due to a (fixed) bug in CPython, you'll need to unset the __PYVENV_LAUNCHER__ variable before activating a venv.

macOS: 'No module named XYZ' errors after running cibuildwheel

cibuildwheel on Mac installs the distributions from Python.org system-wide during its operation. This is necessary, but it can cause some confusing errors after cibuildwheel has finished.

Consider the build script:

python3 -m pip install twine cibuildwheel
python3 -m cibuildwheel --output-dir wheelhouse
python3 -m twine upload wheelhouse/*.whl
# error: no module named 'twine'

This doesn't work because while cibuildwheel was running, it installed a few new versions of 'python3', so the python3 run on line 3 isn't the same as the python3 that ran on line 1.

Solutions to this vary, but the simplest is to install tools immediately before they're used:

python3 -m pip install cibuildwheel
python3 -m cibuildwheel --output-dir wheelhouse
python3 -m pip install twine
python3 -m twine upload wheelhouse/*.whl

macOS: Passing DYLD_LIBRARY_PATH to delocate

macOS has built-in System Integrity protections which limits the use of DYLD_LIBRARY_PATH and LD_LIBRARY_PATH so that it does not automatically pass to children processes. This means if you set DYLD_LIBRARY_PATH before running cibuildwheel, or even set it in CIBW_ENVIRONMENT, it will be stripped out of the environment before delocate is called.

To work around this, use a different environment variable such as REPAIR_LIBRARY_PATH to store the library path, and set DYLD_LIBRARY_PATH in CIBW_REPAIR_WHEEL_COMMAND_MACOS, like this:

Environment variables

CIBW_REPAIR_WHEEL_COMMAND_MACOS: >
    DYLD_LIBRARY_PATH=$REPAIR_LIBRARY_PATH delocate-listdeps {wheel} &&
    DYLD_LIBRARY_PATH=$REPAIR_LIBRARY_PATH delocate-wheel --require-archs {delocate_archs} -w {dest_dir} {wheel}

pyproject.toml

[tool.cibuildwheel.macos]
repair-wheel-command = [
    "DYLD_LIBRARY_PATH=$REPAIR_LIBRARY_PATH delocate-listdeps {wheel}",
    "DYLD_LIBRARY_PATH=$REPAIR_LIBRARY_PATH delocate-wheel --require-archs {delocate_archs} -w {dest_dir} {wheel}"
]

See #816, thanks to @phoerious for reporting.

Windows: 'ImportError: DLL load failed: The specific module could not be found'

Visual Studio and MSVC link the compiled binary wheels to the Microsoft Visual C++ Runtime. Normally, these are included with Python, but when compiling with a newer version of Visual Studio, it is possible users will run into problems on systems that do not have these runtime libraries installed. The solution is to ask users to download the corresponding Visual C++ Redistributable from the Microsoft website and install it. Since a Python installation normally includes these VC++ Redistributable files for the version of the MSVC compiler used to compile Python, this is typically only a problem when compiling a Python C extension with a newer compiler.

Additionally, Visual Studio 2019 started linking to an even newer DLL, VCRUNTIME140_1.dll, besides the VCRUNTIME140.dll that is included with recent Python versions (starting from Python 3.5; see here for more details on the corresponding Visual Studio & MSVC versions used to compile the different Python versions). To avoid this extra dependency on VCRUNTIME140_1.dll, the /d2FH4- flag can be added to the MSVC invocations (check out this issue for details and references).

To add the /d2FH4- flag to a standard setup.py using setuptools, the extra_compile_args option can be used:

    ext_modules=[
        Extension(
            'c_module',
            sources=['extension.c'],
            extra_compile_args=['/d2FH4-'] if sys.platform == 'win32' else []
        )
    ],

To investigate the dependencies of a C extension (i.e., the .pyd file, a DLL in disguise) on Windows, Dependency Walker is a great tool.