Skip to content

Latest commit

 

History

History
161 lines (125 loc) · 5.35 KB

README.rst

File metadata and controls

161 lines (125 loc) · 5.35 KB
Raw

Travis AppVeyor

About

This package is a wrapper around the cibuildwheel tool, and makes it easy to automate building wheels from PyPI source releases rather than for a source repository.

Note that this is intended to be used with packages that require platform-specific wheels. If you have a pure-Python package, you can probably just build a universal wheel with:

python setup.py bdist_wheel --universal

Basic usage

To use autowheel, you should first follow the instructions for setting up cibuildwheel, but instead of running:

pip install cibuildwheel==0.9.4
cibuildwheel --output-dir wheelhouse

You should instead run:

pip install autowheel
autowheel platform --output-dir wheelhouse

where platform is one of macos, linux, or windows. In addition, you should create a file autowheel.yml that contains a file that looks like:

- package_name: mpl-scatter-density
  python_versions:
    '0.1':
      - cp27
      - cp34
      - cp35
      - cp36
      - cp37
    '0.2':
      - cp34
      - cp35
      - cp36
      - cp37

The meaning of python_version is the Python versions that should be build for each package version - but note that you don't need to specify all the package versions - just the ones where the required Python versions change. If the version is not one of the ones listed, the latest one that is equal or less than the required one will be used - in the above example, version 0.1.1 would be built with the same Python versions as 0.1, and 0.3 would be built with the same versions as 0.2.

The way autowheel works is that it will look at all the releases of the package on PyPI that are more recent than the oldest version mentioned in the autowheel.yml file, and for each of them it will determine whether any wheels are missing. If so, then wheel are built for all Python versions specified, and placed in the output directory. To force all wheels to be built even if they already exist, use the --build-existing option:

autowheel platform --output-dir wheelhouse --build-existing

Note that you can list multiple packages inside a single autowheel.yml file, and you can also list the same package multiple times with different configuration if needed.

Options

There are a few options that you can add to the configuration to customize the build process, and which we now describe.

before_build

You can also specify a before_build key with a command that should be run before the wheel is built, e.g.:

- package_name: fast-histogram
  before_build: pip install numpy==1.12.1
  python_versions:
    '0.1':
      - cp27

Note that if you want to pin Numpy, you should take a look at the pin_numpy option instead.

pin_numpy and pin_numpy_min

Wheels that require Numpy as a build-time dependency are typically built against the oldest compatible version of Numpy on each platform and for each Python version. Rather than having to figure this out manually, you can simply set the pin_numpy option as follows:

- package_name: fast-histogram
  pin_numpy: true
  python_versions:
    '0.1':
      - cp27
      - cp35

and autowheel will automatically determine the correct Numpy version to pin against (it does this by determining the version of the oldest available numpy wheels for each platform and python version).

In some cases, this might pick a version of Numpy that is too old for the package you are building, so you can specify an absolute minimum version with the pin_numpy_min option:

- package_name: fast-histogram
  pin_numpy: true
  pin_numpy_min: 1.13.0
  python_versions:
    '0.1':
      - cp27
      - cp35

This means that autowheel will pin Numpy to the oldest compatible version or 1.13.0, whichever is more recent.

test_command and test_requires

An important step when building wheels is to make sure that the built package works properly. You can specify a test command to run after the build using the test_command option:

- package_name: fast-histogram
  test_command: pytest --pyargs fast_histogram
  python_versions:
    '0.1':
      - cp27
      - cp35

In the above case, the --pyargs ensures that we test the installed version of the package rather than the source directory. Note that currently due to the way cibuildwheel works, the tests are run in the same environment as the build environment, so any build-time dependencies installed will still be available (this may change in future).

To install additional dependencies into the test environment (e.g. pytest) or to update dependencies that were installed during the build process, you can use the test_requires option:

- package_name: fast-histogram
  test_command: pytest --pyargs fast_histogram
  test_requires: pytest numpy==1.15.4
  python_versions:
    '0.1':
      - cp27
      - cp35