Random segfault for QEMU builds in GitHub Actions

[Rot127]

Description

This is not so much an issue of cibuildwheel (I think), except you should be aware of it.

GCC builds segfault on the AArch64 musllinux 1.2, manylinux 2014 and manylinux 2.28.

I couldn't pin-point a possible GCC issue to it. But it is odd that it appears in a stable version.
Maybe: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118583
Haven't tested it yet though.

Feel free to close this if you see it as out of scope. But I wanted to drop it here in case people search for it in this context. Or others already found a solution.

Runs:

manylinux 2.28
https://github.com/capstone-engine/capstone/actions/runs/13012237284/job/36292511727?pr=2614

manylinux 2014
https://github.com/capstone-engine/capstone/actions/runs/13011576391/job/36290786445?pr=2614

musllinux 1.2
https://github.com/capstone-engine/capstone/actions/runs/13011576391/job/36290786852?pr=2614

Build log

No response

CI config

No response

[henryiii]

It's a bug in emulation. You can either pin the QEMU version, or better yet, use ubuntu-24.04-arm or ubuntu-22.04-arm instead of emulation, which is ~10x faster too.

See #2256. I'll copy @joerick's answer here for posterity:

I believe you're hitting a bug around setup-qemu-action. I didn't see it myself but @mayeut just pushed this fix to our repo-

4538497

The referenced issues are docker/setup-qemu-action#188 and tonistiigi/binfmt#215. @mayeut's workaround is the following in a github workflow:

    - name: Set up QEMU
      if: runner.os == 'Linux'
      uses: docker/setup-qemu-action@v3
      with:
        # This should be temporary
        # xref https://github.com/docker/setup-qemu-action/issues/188
        # xref https://github.com/tonistiigi/binfmt/issues/215
        image: tonistiigi/binfmt:qemu-v8.1.5

[nightlark]

I missed it initially, but buried in the binfmt issue that was linked above there is discussion about the cause being related to building qemu with --disable-pie and aslr with the kernel the GHA runners ship (and qemu 8.1+ having a patch related to the segfault): tonistiigi/binfmt#215 (comment)

[e-dant]

Related? https://github.com/e-dant/watcher/actions/runs/13093964783/job/36534092020#step:4:511

Command ['which', 'python'] failed with code 139.

[mayeut]

Related?

@e-dant, that's the same issue: random segfaults with QEMU.