On Wed, Sep 10, 2025 at 2:01 PM Charles R Harris via NumPy-Discussion < numpy-discussion@python.org> wrote:
> > > On Wed, Sep 10, 2025 at 2:45 AM matti picus via NumPy-Discussion < > numpy-discussion@python.org> wrote: > >> On Tue, Sep 9, 2025 at 10:11 PM Ralf Gommers via NumPy-Discussion >> <numpy-discussion@python.org> wrote: >> > >> > Hi all, >> > >> > GitHub Actions has deprecated the macos-13 image [1], which was the >> last image to support Intel x86-64 runners. As a result, we won't be able >> to use our current wheel build setup. >> > >> > A quick sampling of what others are doing with their support policies: >> CPython dropped it to Tier 2 [2], Anaconda dropped support [3], conda-forge >> hasn't decided anything yet [4], PyTorch dropped support already 1.5 years >> ago [5], Numba is dropping wheels [6]. >> > >> > Cheers, >> > Ralf >> >> Azure pipelines (use by conda-forge) still offers macos x86_64 >> machines, but they are not Trusted Publishers to PyPI. So if we choose >> to use them to create wheels in the numpy-release repo, we could >> continue to manually upload those wheels. This has some security >> risks. In issue 29178 [0] Ralf laid out a security roadmap that >> includes Trusted Publishing as a goal. Personally, I feel we should be >> one of the last libraries to stop publishing binary wheels, and >> perhaps in the security risk vs. usability balance we could choose to >> manually upload wheels for another year or so. >> Matti >> >> [0] https://github.com/numpy/numpy/issues/29178 > > > I agree it would be nice to support x86_64 for about a year more, but I > don't see how we are going to do it. We may need to leave it to third > parties, as we do other "exotic" hardware support. Apple has been the prime > mover here, the rest of us are just adapting. > > Speaking of exotic hardware, are we going to be able to publish PPC > wheels? > That's a separate decision, I'm still on the fence. Since yesterday we now have a regular CI job (https://github.com/numpy/numpy/pull/29212), which is quite nice. I think having more regular CI jobs for more niche platforms like this one and the ones we have for FreeBSD and under QEMU for some other platforms like 32-bit Arm and RISC-V are very useful in making NumPy more reliable. That said, for wheels we'd need to cross-compile in the `numpy-release` repo and rely on regular CI in the main repo to ensure the test suite passes. Which is fine, but it's also extra work and extra CI load, so it's a "is it worth it" question. That said, I'm hoping we will have proper cross compilation support very soon, because PEP 739 support in Meson got merged. So it may actually not be that high-effort after that. The question then is still whether we want to publish wheels on PyPI, since it's a long-term commitment for a limited set of users - hard to estimate how many there are, but less than 1% of the total user base almost certainly. Then again, we did add Windows arm64 wheels, which have way fewer users than PowerPC. So it's always a judgement call. > What is PyPI policy on certifying trusted publishers? > There can be multiple in principle, but the only ones that seem supported as of now are GitHub Actions, Google Cloud, ActiveState, and GitLab CI/CD: https://docs.pypi.org/trusted-publishers/using-a-publisher/. We'd strongly prefer to have only one, because if there are multiple there's awkward coordination issues between different systems. Cheers, Ralf
_______________________________________________ NumPy-Discussion mailing list -- numpy-discussion@python.org To unsubscribe send an email to numpy-discussion-le...@python.org https://mail.python.org/mailman3//lists/numpy-discussion.python.org Member address: arch...@mail-archive.com
