On Fri, Nov 6, 2026 at 2:29 AM Nathan via NumPy-Discussion < [email protected]> wrote:
> Thanks, that extra context helps. > > I took a stab at updating our use of Mac CI runners to improve throughput: > https://github.com/numpy/numpy/pull/33227 > > On Thu, Nov 5, 2026 at 6:06 PM Malcolm Smith <[email protected]> wrote: > >> Hi Nathan, >> >> The method of consumption is basically the same as on desktop platforms: >> a developer runs pip or uv, and it downloads the wheel from a repository >> and installs it into an environment. This can be done natively on a mobile >> device using apps like Termux which provide a fairly conventional >> development environment. Or it can be done on a desktop machine using the >> other projects I mentioned, which build the environment into a mobile app >> which can be distributed on an app store to end users who might not even >> know they're using Python. >> >> Right now each of those projects is maintaining a limited set of binary >> Python packages that they support on mobile. Every new package, and every >> new version of each package, takes up our time. With 100 or so packages >> currently supported, even updating them once per year is about as much as >> we can keep up with, but our users are constantly asking for more. And to >> make using Python on mobile realistic for most applications, I think we'd >> need to push that number closer to 1000. >> >> The only feasible way to achieve that is to remove ourselves as the >> bottleneck, and have each upstream project release wheels for mobile on >> PyPI in the same way as all the other platforms. >> >> -- Malcolm Smith >> >> On Wed, 4 Nov 2026 at 19:51, Nathan via NumPy-Discussion < >> [email protected]> wrote: >> >>> Can you explain a little more what the concrete upside is here? For >>> context: I've never built a phone app, let alone tried to consume a ios or >>> android wheel. >>> >>> How would a developer targeting ios or android consume the PyPI wheels >>> and why is what they currently do a big pain for downstream? >>> >>> Building new wheels is a concrete ongoing commitment from us. >>> >>> One thing I've noticed is that ths ios builds in CI are often blocked on >>> available Mac GitHub runners but that's a very minor issue. Still, it'd be >>> nice to figure out how to run fewer Mac CI jobs or exploit parallelism more >>> efficiently. >>> >>> On Mon, Nov 2, 2026 at 6:26 AM Malcolm Smith <[email protected]> wrote: >>> >>>> Hi everyone, >>>> >>>> As you may remember, Android and iOS support was merged into NumPy a >>>> few months ago, and both of them were added to NEP 57 as Tier 3 platforms. >>>> As far as I'm aware, they've both been running reliably in CI since then, >>>> so I'd like to propose moving them to Tier 2 and releasing their wheels on >>>> PyPI starting with NumPy 2.6 next month. >>>> >>> I believe that those CI jobs have been very stable, so from that perspective this has gone well so far. And in the Tier 3 promotion (thread <https://mail.python.org/archives/list/[email protected]/message/BQX3WGZI23OQJSNQSVBS5TAPUPNIROUR/>) we also touched on Limited C API support, and that should all work so that's an option as soon as NumPy supports it - and is an option today for most other projects in the ecosystem. I believe that that's the most critical thing for whether to accept a new platform with low usage numbers - we cannot keep adding 6 wheels (and hence 6 CI jobs for nightlies/releases that all can fail) per platform, but 1 is much more palatable. As long as we do have contributors who are platform experts and volunteer to keep helping in case there are problems (which we do have here with Malcolm and Russell), I'm mildly in favor of accepting thing proposal. Re macOS runner load: we should be able to drop macOS x86-64 support next year, which more than halves the load. I'd actually be fine with doing that after the 2.6.0 release, because those jobs are quite unstable and hence a maintenance burden. Cheers, Ralf > >>>> This will allow users of Termux, Chaquopy, BeeWare and Kivy to obtain >>>> up-to-date versions of NumPy, without each of those projects needing to >>>> maintain their own builds as they've been doing until now. >>>> >>>> Let me know if you have any questions. >>>> >>>> Thanks, >>>> Malcolm >>>> _______________________________________________ >>>> NumPy-Discussion mailing list -- [email protected] >>>> To unsubscribe send an email to [email protected] >>>> https://mail.python.org/mailman3//lists/numpy-discussion.python.org >>>> Member address: [email protected] >>>> >>> _______________________________________________ >>> NumPy-Discussion mailing list -- [email protected] >>> To unsubscribe send an email to [email protected] >>> https://mail.python.org/mailman3//lists/numpy-discussion.python.org >>> Member address: [email protected] >>> >> _______________________________________________ > NumPy-Discussion mailing list -- [email protected] > To unsubscribe send an email to [email protected] > https://mail.python.org/mailman3//lists/numpy-discussion.python.org > Member address: [email protected] >
_______________________________________________ NumPy-Discussion mailing list -- [email protected] To unsubscribe send an email to [email protected] https://mail.python.org/mailman3//lists/numpy-discussion.python.org Member address: [email protected]
