Hello,
As @webknjazreported, GitHub will be dropping their last Intel Mac image. Brownouts will begin September 1, so we need to make the job non-blocking by then.
Cirrus and Circle CI already dropped Intel Mac.
This means that we’ll likely need to move Intel Mac to Tier 2 in PEP 11.
Another option would be to run the tests with Rosetta, but setting that up would probably be too much work for the benefit.
See the issue for more details.
What makes me a bit worried is that Tier 2 requires 2 core dev contacts. How many core devs still use an Intel Mac?
We do have buildbots, thanks to Matt Billenstein and @itamaro.
I don’t have a stake in this (and I don’t own a Mac) but I wonder if “x64 Python running on ARM hardware on OSX with Rosetta” might actually be a more important platform than “x64 Python running on x64 on OSX”. Just because people will be using it to run outdated libraries. So the suggestion of using GitHub’s more recent images and Rosetta might not be terrible.
I’m only worried that native vs. virtualized macOS envs might have different surrounding ecosystems.
Though, I see your point. Ever since macOS 12.3 dropped system-provided Python (2.7), the end-users always end up relying on CPython provided by non-Apple vendors. Be it python.org, Homebrew, uv, or pyenv. So with this in mind, whatever the upstream Python decides would match whatever runtime the users will get.
I can imagine, however, that dropping x86_64 may force certain frameworks to prolong having older CPythons in their text matrixes, past their respective upstream EOL dates.
Thinking of the real-world use-cases possibly connected to using x86_64 Macs, one such case might be the Ansible ecosystem, where the community maintaining shareable automations might want to support managing older machines for longer, past their expiration date. Another use of the old Macs I can think of is dual-boot — Boot Camp only supports Intel hardware.
That said, if Rosetta2 provides a sufficiently equivalent environment (I don’t have any experience with it), that’s probably the way to go.
I also still have a 2018 Intel mac mini, but recently made the upgrade to an M4 because of the rumors (confirmed?) that macOS 26 will not support this machine. Dropping to Tier 2 makes sense to me too.
I have an 2018 mini, and will keep that running for testing other projects anyway.
The more annoying cutoff will be when Apple stops shipping x86_64 target support in their compilers and SDKs, although that might be years out at the moment (macOS 26 will be the last version that runs on some Intel Macs, macOS 28 will drop support for Rosetta[1])
Building/testing with Rossetta shouldn’t be too hard if Rossetta is installed in GitHub’s runners, but is more complexity that would need handholding at times. So +1 for dropping Intel macOS to tier 2.
except for some special cases that are not relevant for this discussion. ↩︎
Starting August 15, 2025, Anaconda will stop building new packages for Intel Mac computers (osx-64). This change affects users who are currently using Intel Mac packages, whether on actual Intel hardware or Apple Silicon Macs running in emulation mode.
For completeness, conda-forge is still building for osx-64 on macOS >=10.13 at the moment. So far, we’ve been only discussing to bump the lower bound to 11.0, and I wouldn’t have expected an end of us building osx-64 packages anytime soon, but if GHA / Azure Pipelines ends up dropping Intel Mac agents, we may be forced to.
Ironically, there are still no native osx-arm64 runners, though they’re scheduled for this Q3 apparently. It looks like it’ll be a very thin window to change our infrastructure over from osx-64 to osx-arm64 runners. In principle, we should be able to cross-compile osx-arm64 from osx-64, just as we’ve been doing it the other way around for several years now since the first M1. I’m pretty sure this will be attempted; whether we keep osx-64 will depend on how much friction that change causes, and whether people put up with the required effort to keep it going or not.
Update: the PR has been merged, and Intel macOS (aka x86_64-apple-darwin) is now tier 2:
We’re still testing macos-13 on the CI, and I suggest we keep it there until GH pulls the plug (some time between 1st September and 14th November), as it’ll help us surface failures earlier.
But when it does go away, here’s a PR that’s ready to merge and minimise the disruption: