Wheels are built and pushed to PyPI successfully for all OSs and architectures: 1, 2. As before, we're utilizing 2 build systems: Cirrus CI (macos_arm64 and linux_aarch64) and GHA (macos_x86_64, linux_x86_64, and win_amd64). While GHA supports building natively on Apple Silicon via macos-latest-xlarge runner, this feature is limited to Teams/Enterprise accounts AFAIK. Besides, Cirrus CI provides substantially lower free credits relative to GHA, so there will be instances where the CI/CD for arm64 will fail to execute due to insufficient credits. Therefore, as a stopgap, I've also provisioned a self-hosted macOS VM based on Cirrus's Tart virtualization that can run on Apple Silicon via wheels-macosx_arm64.yml. For this last bit to work, a self-hosted runner needs to be initialized via GitHub while I donate CPU time on my M1 MacBook Air as and if the need arises.
Pushing to PyPI is confirmed working using multiple build systems, so if the secrets.PYPI_API_TOKEN repository secret is configured it should be seamless. For Cirrus CI, once access to ANTsX/ANTsPy has been granted at https://cirrus-ci.com/, the runner should automatically read the config from .cirrus.yml.
In terms of PyPI storage, I believe we're at the max 10GB for the project, so upgrade requests should be in order to support another version or two.
On a related note, if there's interest, I can test pushing the multi-arch image (amd64/ arm64) directly to Docker Hub - build works already through Cirrus CI. The current workflow (ci-docker.yml) only supports amd64. An access token or access delegation would be needed to accomplish that.
Wheels are built and pushed to PyPI successfully for all OSs and architectures: 1, 2. As before, we're utilizing 2 build systems: Cirrus CI (
macos_arm64
andlinux_aarch64
) and GHA (macos_x86_64
,linux_x86_64
, andwin_amd64
). While GHA supports building natively on Apple Silicon viamacos-latest-xlarge
runner, this feature is limited to Teams/Enterprise accounts AFAIK. Besides, Cirrus CI provides substantially lower free credits relative to GHA, so there will be instances where the CI/CD forarm64
will fail to execute due to insufficient credits. Therefore, as a stopgap, I've also provisioned a self-hosted macOS VM based on Cirrus's Tart virtualization that can run on Apple Silicon viawheels-macosx_arm64.yml
. For this last bit to work, a self-hosted runner needs to be initialized via GitHub while I donate CPU time on my M1 MacBook Air as and if the need arises.Pushing to PyPI is confirmed working using multiple build systems, so if the
secrets.PYPI_API_TOKEN
repository secret is configured it should be seamless. For Cirrus CI, once access to ANTsX/ANTsPy has been granted at https://cirrus-ci.com/, the runner should automatically read the config from.cirrus.yml
.In terms of PyPI storage, I believe we're at the max 10GB for the project, so upgrade requests should be in order to support another version or two.
On a related note, if there's interest, I can test pushing the multi-arch image (
amd64
/arm64
) directly to Docker Hub - build works already through Cirrus CI. The current workflow (ci-docker.yml
) only supportsamd64
. An access token or access delegation would be needed to accomplish that.Cheers!