release CI fails to build for aarhc64 targets

[2bndy5]

In the release CI, it fails to build for Python v3.9.x on aarch64 (using qemu and a docker image).

(unhelpful) CI build/release logs

CPython 3.9 manylinux aarch64

Setting up build environment...
  
      + /opt/python/cp39-cp39/bin/python -c 'import sys, json, os; json.dump(os.environ.copy(), sys.stdout)'
      + which python
      + which pip
                                                              ✓ 0.57s
Building wheel...
  
      + rm -rf /tmp/cibuildwheel/built_wheel
      + mkdir -p /tmp/cibuildwheel/built_wheel
      + python -m pip wheel /project --wheel-dir=/tmp/cibuildwheel/built_wheel --no-deps
  Processing /project
    Installing build dependencies: started
    Installing build dependencies: finished with status 'done'
    Getting requirements to build wheel: started
    Getting requirements to build wheel: finished with status 'done'
    Preparing metadata (pyproject.toml): started
    Preparing metadata (pyproject.toml): finished with status 'done'
  Building wheels for collected packages: pyrf24
    Building wheel for pyrf24 (pyproject.toml): started
    Building wheel for pyrf24 (pyproject.toml): finished with status 'error'
    error: subprocess-exited-with-error
    
    × Building wheel for pyrf24 (pyproject.toml) did not run successfully.
    │ exit code: -11
    ╰─> [0 lines of output]
        [end of output]
    
    note: This error originates from a subprocess, and is likely not a problem with pip.
    ERROR: Failed building wheel for pyrf24

Still investigating, but it will be difficult.

FYI, Python v3.8 and v3.9 have hit End-of-Life. They aren't receiving any new updates (security or otherwise). It would be simple to bump the minimum required python version for this pkg:

pyRF24/pyproject.toml

Line 49 in a3f760f

requires-python = ">=3.7"

But piwheels still distributes v3.9 builds (for bullseye installs).

[2bndy5]

pybind11 will drop support for Python v3.7 in next release. Currently, we are pinned to pybind11 v2.13.6 (#96) which allegedly still supports Python v3.7+.

[2bndy5]

Well I had to build python v3.9.21 from tarball source. I didn't run into any problems though.

I did have to add the Python-v3.9.21/Include directory so pybind11 could find the right Python.h to #include. But doubt that is the same problem observed in CI because the aarch64 build uses a docker image.

Still investigating...

[2bndy5]

Some progress here

I successfully built the package for the aarch64 platform locally (in Ubuntu) using QEMU, Docker, and the cibuildwheel package. So, there must be something wrong with the env used in the CI workflow.

After running the CI maunually, I found 2 different errors:

• the compiler seg faults. As usual in this situation, there is no indication as to why it seg faults. Compiler version reported as GNU 10.2.1. Locally, the build completes as expected using GNU 13.3.0.
• The docker image used in QEMU may not have git installed. Although, this problem didn't occur after I disabled the automatic caching of the qemu docker image.

[2bndy5]

I think I found the problem: tonistiigi/binfmt#215. Its a problem with the actual emulator with respect to aarch64 containers on amd64 hosts.

I think the solution is to use a newer qemu release. Currently the latest tag resolves to a 2-year old release.

[2bndy5]

Yep. Bumping the QEMU version to v8.5.1 resolved the seg fault issues. I suspect this was recently brought on by GitHub's upgrades to the ubuntu-latest runner (which will now use Ubuntu 24.04).