Corridor pypi mirror

This is a mirror with packages from pypi - making it easier to install libraries and dependencies for Corridor Platforms.

Usage:

pip install <pkgname> --extra-index-url https://corridor.github.io/pypi

The package/version/target list is maintained in wheel-builds.json.

Why ?

Pypi is great, but sometimes packages that we need are not available there.

For example:

1. Some libraries support older python versions, but wheels are not present
2. Some older versions of libraries, don't have wheels for new python versions

Package Index Creation

The GitHub Actions workflow:

1. Expands wheel-builds.json into a build matrix
2. Reuses a compatible wheel from its GitHub Release, when one already exists
3. Builds only missing wheels with the requested CPython version
4. Publishes new wheels as immutable GitHub Release assets
5. Creates a PEP 503 package index linking to those assets
6. Deploys the result to GitHub Pages

Release assets are the persistent wheel store; GitHub Actions artifacts are used only to pass wheels between jobs in one workflow run. Existing release assets are never overwritten. If a wheel must be rebuilt and its bytes change, publish it under a new package version instead of replacing the existing asset.

Add one JSON object when a package, version, or target is needed. Commits to master rebuild and redeploy the complete index.

Linux manylinux_2_28_x86_64 wheels are built inside the official PyPA Docker image, which supplies CPython 3.11–3.14 and auditwheel. macOS ARM64 wheels are built natively on a macos-15 ARM64 runner; auditwheel is Linux-only. PySpark produces a universal py2.py3-none-any wheel, so duplicate builds for different Python versions or targets collapse to one file in the published index.

The configuration uses manylinux_2_28_x86_64 because that is the supported PyPA image target. A manylinux_2_26_x86_64 entry is rejected by matrix validation until a suitable image and auditwheel policy are available.