Switch autobuild OS to ARM runner

[ann0see]

Short description of changes

Switches the build to ARM runners if an arm build is produced

CHANGELOG: Build: Switch Linux ARM builds to be built on arm runners

Context: Fixes an issue?

Fixes: #3466

Does this change need documentation? What needs to be documented and how?

No

Status of this Pull Request
Tested on debian arm64 on a raspberry pi. armhf and arm64 both run. I'd like to have an independent test too (best on raspbian).

What is missing until this pull request can be merged?

External testing

Checklist

• I've verified that this Pull Request follows the general code principles
• I tested my code and it does what I want
• My code follows the style guide
• I waited some time after this Pull Request was opened and all GitHub checks completed without errors.
• I've filled all the content above

[ann0see]

ARM runners are considered public beta. I believe that if we don't remove the possibility to build on x86, it would be ok to merge this. But I'd also like to hear a general opinion.

[softins]

I won't have time to test these builds for several days yet.

I've just looked at the build logs, and notice that native compilation only works for amd64 and arm64 architectures. Building armhf is still cross-compilation, but now on an arm64 runner. I assume that's expected.

[ann0see]

Building armhf is still cross-compilation, but now on an arm64 runner. I assume that's expected.

Yes. That's expected.

[softins]

Armhf build loads and runs ok on my Pi 4.

[pljones]

What happens if "${APT_ARCH}" is arm64 and "${TARGET_ARCH}" is "amd64", for example? Should this explicitly error before this point unless"${APT_ARCH}" == "amd64"? (Rather than just not doing the copy.)

[softins]

That's a very good question. The present logic here seems to assume that cross-compilation is only done on an amd64 runner. That's no longer the case: we explicitly cross-compile from arm64 to armhf on an ARM runner. So I wonder whether this arch duplication is still needed in that case?

It would be good to get a reference for where this technique came from and why/when it's needed. Is @hoffie still around?

[softins]

OK, I think I understand it. It seems purely a method of specifying repo URLs independently for each architecture.

It seems most Ubuntu repos, such as the UK mirror only contain amd64 and i386, so the lines in /etc/apt/sources.list will only be able to find packages for those architectures.

When we want to cross-compile to non-Intel architectures, we need to be able to fetch packages from Ubuntu Ports instead. All the sed stuff is just setting up the routing for fetching those architectures, and also needs to prevent going there for amd64 and i386, which the Ports repo doesn't contain.

There is an example Gist here that shows a sources.list set up for cross-compilation in that way. The sed stuff in our repo just uses sources.list.d/ to organise the deb records for fetching non-native architectures.

I haven't had time yet to work out whether we need to do this for arm64 runners compiling for armhf. Maybe we don't.

[pljones]

Approving on the basis that we'll find out if it breaks when it breaks.