⚠️ This repo is slightly outdated, and these days, there are better examples.

For auto-deployment examples, take a look at the cibuildwheel /examples folder, particularly the Github Deploy example.

• Add an env section to your .travis.yml

  env:
    global:
      - TWINE_USERNAME=__token__
      # Note: TWINE_PASSWORD is set to an API token in Travis settings
  

  Install cibuildwheel and twine in your install section

  install:
    - python -m pip install twine cibuildwheel==x.y.z
  

  Build in the script section

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

  Finally, upload if the build was successful

   after_success:
     - if [[ $TRAVIS_TAG ]]; then python -m twine upload wheelhouse/*.whl; fi
  

  Check this repo's .travis.yml as an example.

• Generate a PyPI API token

• In the Travis web UI, go to your project settings and add the environment variable TWINE_PASSWORD, set to your new PyPI API token.

• Add this env to your appveyor.yml

  environment:
    TWINE_USERNAME: __token__
    # Note: TWINE_PASSWORD is set in Appveyor settings
  

  Add this upload step to the build_script:

  build_script:
    - pip install cibuildwheel==x.x.x
    - cibuildwheel --output-dir wheelhouse
    - >
      IF "%APPVEYOR_REPO_TAG%" == "true"
      (
      python -m pip install twine
      &&
      python -m twine upload wheelhouse/*.whl
      )
  

  Check this repo's appveyor.yml as an example.

• Generate a PyPI API token

• In the Appveyor UI, add your new API token as TWINE_PASSWORD (click Settings > Environment > Add Variable...). Make sure to mark it as private!

• Bump the version number in setup.py and anywhere else it occurs (I use bumpversion for this)
• Commit these changes, tag that commit, and push to Github (don't forget to push the tag! git push --tags). Your wheels will start building.
• Locally, build a source distribution with rm -rf dist && python setup.py sdist
• Upload the source distribution using twine upload dist/*.tar.gz

Your wheels will build in Travis/Appveyor and push to PyPI when ready.