ENH: Add armv7l wheel builds (manylinux/musllinux)#31348
Conversation
fe30834 to
0c2232b
Compare
|
I'd note that the RPi5 units, and the CM5 modules, can both use the manylinux_aarch64 wheels that are currently on PyPI. |
|
I'm pretty sure that CM4 and RPi4 would also be able to use the manylinux_aarch64 wheels as the tech specs state "Broadcom BCM2711, Quad core Cortex-A72 (ARM v8) 64-bit SoC @ 1.8GHz" |
Yes that is true, but we are stuck at 32-bit and cannot easily upgrade, as they are already deployed |
|
I have now opened up an issue: #31351 |
|
@Lauszus did you test the new CI jobs on your fork, and if so can you link to a passing log? Note that it's normal to develop/debug CI jobs on your fork first, because you can run just the relevant job, rather than triggering O(75) jobs on a PR. |
|
It looks like for regular armv7l CI coverage, we already have that: numpy/.github/workflows/linux.yml Line 170 in 3bea241 It was just a little harder to dig up because of the So my read is that this doesn't increase coverage, unless we want to consider armv7l on Musl - but does anyone use that, and does it turn up any new issues now that it's run for the first time (see my question about results above)? |
PR summary
This PR adds support for building armv7l (Raspberry Pi) wheels.
Addresses the feature request in gh-31351. Please also see my corresponding numpy release PR: numpy/numpy-release#45
First time committer introduction
My name is Kristian and currently work at https://candela.com/ where we use numpy for various Python applications. Some of these applications are running on a Raspberry Pi compute module 4. For now we have been building numpy from source in CI by extending the official manylinux Docker images.
I thought it would be nice to build it upstream instead to speed up the CI, but also I think it will benefit a lot of people using Raspberry Pi for various hobby projects.
AI Disclosure
I used Claude Sonnet 4.6 to fork and push the commit, but I told it exactly which lines to change.