Support windows-11-arm

[ntkme]

GitHub Actions recently added new windows-11-arm runner for OSS: https://github.blog/changelog/2025-04-14-windows-arm64-hosted-runners-now-available-in-public-preview/

• Closes Request: Support Windows on Arm64 #600

This PR and the following adds support for windows-11-arm:

• Support clangarm64 environment setup-msys2-gcc#21

[ntkme]

The RubyInstaller-Devkit package consists of the LLVM based C and C++ compilers which run natively on ARM. This makes the package almost twice as big than the x64/x86 versions. In contrast the MSYS2 toolset, which provides Unix tools like make and bash, is not yet ported to ARM. It is delivered as x64 code, running on the Prism emulator of Windows-11. That’s why only the RubyInstaller version without Devkit will work on Windows-10.

setup-ruby does not use devkit from rubyinstaller so we will have to replicate a similar setup with https://github.com/ruby/setup-msys2-gcc.

@MSP-Greg @larskanis I can start working on a PR for that, but I'd like to hear your thoughts.

[larskanis]

Thanks to notify me about the new Windows on ARM64 runners!

setup-ruby does not use devkit from rubyinstaller so we will have to replicate a similar setup with https://github.com/ruby/setup-msys2-gcc.

This shouldn't be too difficult. Since the same MSYS2 x64 executables are used on arm64 as on x64, the msys2.7z file should be reusable. Only the clangarm64 environment needs to be packaged into a 7z file equal to the mingw64 and ucrt64 environments. I think packaging these clangarm64 files should even work on a x64 host. See https://www.msys2.org/wiki/arm64/

There is no need to select distinct packages regarding gcc on x64 and clang on arm64. The gcc packages are wrappers for the corresponding clang packages on arm64, so that the same package list can be installed per pacman. Only the different prefix mingw-w64-clang-aarch64 vs. mingw-w64-ucrt-x86_64 has to be respected. That's why the rubyinstaller package list didn't change with the introduction of WoA.

[ntkme]

I think packaging these clangarm64 files should even work on a x64 host.

Not that easy. The way setup-msys2-gcc works today is that it uses the C:/msys64 preinstalled on the x64 windows runner, and only pack the "difference" to create a minimal package for these runner. However, the windows-11-arm runner does not have C:/msys64 preinstalled, so an "incremental" package simply won't work because all the base files are missing. We will have to provide two different packages, the existing "incremental" package for x64 and a new "complete" package for arm64. Currently testing that in my own fork.

[ntkme]

This PR is ready for review. The tests are failing because other linked PRs are required. Also there is a chicken-and-egg problem with this PR and ruby/setup-msys2-gcc#21, which has a few ways to get worked around.

The ways I get it solved in my own forks is that:

1. Use setup-ruby and setup-msys2-gcc changes as is to first generate the msys2 package for windows-11-arm. At this stage the clangarm64 gcc package will fail, but we will get a working msys2-windows-11-arm package.
2. Use the setup-ruby change, but slightly modify it to skip the gcc setup only on windows-11-arm. - This allows us to build clangarm64 gcc package.
3. Now we have a working state with both msys2-windows-11-arm and clangarm64 gcc.

[eregon]

However, the windows-11-arm runner does not have C:/msys64 preinstalled

And there is no plan to add it by GitHub?

[eregon]

This looks great, but of course we'll need to wait that the related PRs are ready and approved as well

[ntkme]

And there is no plan to add it by GitHub?

It's documented here: https://github.com/actions/partner-runner-images/blob/main/images/arm-windows-11-image.md#omitted-software - I think this essentially means they don't plan to add it.

[eregon]

It seems worth asking (via issue/discussion) or searching issues, that list includes things that probably don't exist on windows-arm64, but MSYS2 does work there so not clear why it's not included to me.

[ntkme]

It seems worth asking (via issue/discussion) or searching issues, that list includes things that probably don't exist on windows-arm64, but MSYS2 does work there so not clear why it's not included to me.

actions/partner-runner-images#88

Someone already asked, and it looks like they don't like the fact that msys2 for windows arm64 has hybird x64 emulation and arm64 native...

[ntkme]

I added a commit in this PR to address the circular dependency on this PR and ruby/setup-msys2-gcc#21.

The plan of action is:

1. Merge this PR as is. Two tests will fail temporarily.
2. Merge Support clangarm64 environment setup-msys2-gcc#21. Wait for build to complete.
   • Sample run from my fork: https://github.com/ntkme/setup-msys2-gcc/actions/runs/14663140853
3. Revert the commit Temporarily disable installGCCTools on windows arm64 to solve chicken-and-egg issue. Test will pass.
   • Sample run from my fork: https://github.com/ntkme/setup-ruby/actions/runs/14662729309

[MSP-Greg]

@ntkme

Today I added PR's to my forks of setup-ruby and setup-msys2-gcc, and generated the two 7z files needed for arm64.

So, they now exist in the ruby/setup-msys2-gcc release.

You should be able to remove the 'Temporarily disable installGCCTools' commit and CI should pass, as it did in my fork.

After this is accepted, we can switch the 'setup-msys2-gcc' PR back to using ruby/setup-ruby.

Thanks again,

Greg

[ntkme]

You should be able to remove the 'Temporarily disable installGCCTools' commit and CI should pass, as it did in my fork.

Thanks, this greatly simplifies the operation here.

After this is accepted, we can switch the 'setup-msys2-gcc' PR back to using ruby/setup-ruby.

That PR is already using ruby/setup-ruby as I found out that I can do the workaround in this PR to disable the gcc setup... But now you have uploaded the achieve needed, no workaround is needed anymore, except for that this PR to be merged before the 'setup-msys2-gcc' PR.

[eregon]

Thank you for all your work on this!