chore(deps): Bump pypa/gh-action-pypi-publish from 1.14.0 to 1.14.2

[dependabot]

Bumps pypa/gh-action-pypi-publish from 1.14.0 to 1.14.2.

Release notes

Sourced from pypa/gh-action-pypi-publish's releases.

v1.14.2

🛠️ Urgh… Another release!? Again? Explain yourself!

Looking at the diff, you'll only witness updates across the dependency tree. That's it! It's not a security fix or anything like that even, no. But you'll want this update.

[!tip] So what most people will find useful is @​takluyver💰's update of Twine to v7 that we use internally (#416). This version will let them upload their sdists and wheels containing core packaging metadata v2.5 to (Test)PyPI.

🧐 Tell me why..

TL;DR non-pure-python projects with C-extensions tend to have dozens (sometimes hundreds) wheels to upload to PyPI per release. They are often quite big and take time to transfer over the network. People started noticing problems and coming up with DIY sharding workarounds like aio-libs/aiohttp#13226 around July 23. On this date, projects with a good amount of bytes to publish would start getting timeouts 5 minutes after the PyPI publishing job begun. The same job that worked just fine before.

I had to start pinging upstream library and ecosystem people, on GitHub and privately, to start making sense of what was happening. Eventually, we collectively concluded that GitHub must've shortened the lifetime of their OIDC identity — it seems to have used to be 10 minutes long (at some point in the past) and is now 5 minutes, apparently. It's not documented clearly, and we have not been able to get any clarity by attempting to contact GitHub through private channels, using personal connections.

Over the course of investigation, @​facutuesca💰 found and fixed a related underlying cache invalidation bug in sigstore/sigstore-python#1838, which he then coordinated propagation through the dependency chain updates in sigstore-python, pypi-attestations, gh-action-pypi-publish and gh-action-sigstore-python.

Mike's also discovered that Sigstore's Rekor slowdown seems to have become the main contributing cause of the last week's incident. He's collected some data to support this claim: https://publishing-five-minute-timeout.tiiny.site.

🫶 New Contributors

• @​davidbrochart made their first contribution in #415
• @​takluyver made their first contribution in #416

🪞 Full Diff: pypa/gh-action-pypi-publish@v1.14.1...v1.14.2

🧔‍♂️ Release Manager: @​webknjaz 🇺🇦

🙏 Special Thanks to @​davidbrochart💰 and @​Dreamsorcerer💰 for turning my attention (in #415 and in private) to the newly surfaced corner case in GitHub's behavior that only affected a narrow category of projects while many others remained blissfully unaware. @​bdraco💰 came up with a DIY sharding workaround for aiohttp that served as a demo for other projects. @​miketheman💰 confirmed the Warehouse-side details. Also, @​jku💰 and @​woodruffw💰 helped work through, review and release the Sigstore ecosystem upstream libs.

💬 Discuss on Bluesky 🦋, on Mastodon 🐘 and [on GitHub][release discussion].

[![GH Sponsors badge]][GH Sponsors URL]

... (truncated)

Commits

• dc37677 Merge pull request #417 from trail-of-forks/ft/bump-deps
• 8b2f234 Bump pypi-attestations and sigstore
• 78b72db Merge pull request #416 from takluyver/twine-v7
• 92f4d2a Update twine to v7
• ba38be9 Merge pull request #408 from adisivaprasad/bump-setup-python-v6
• a6c5088 Bump actions/setup-python from v5.6.0 to v6.2.0
• See full diff in compare view