Add Windows on arm64 as a supported platform

[chalcolith]

Allow building on arm64 Windows

[ponylang-main]

Hi @chalcolith,

The changelog - added label was added to this pull request; all PRs with a changelog label need to have release notes included as part of the PR. If you haven't added release notes already, please do.

Release notes are added by creating a uniquely named file in the .release-notes directory. We suggest you call the file 325.md to match the number of this pull request.

The basic format of the release notes (using markdown) should be:

## Title

End user description of changes, why it's important,
problems it solves etc.

If a breaking change, make sure to include 1 or more
examples what code would look like prior to this change
and how to update it to work after this change.

Thanks.

[SeanTAllen]

@chalcolith https://github.com/ponylang/ponyup/blob/main/ponyup-init.ps1 should be updated to.

[SeanTAllen]

The windows corral for arm appears to hang often.

[chalcolith]

I tried adding my local arm64 windows machine as a self-hosted runner. The crypto and net_ssl libs built much faster, so maybe the existing windows-11-arm machine is not very powerful? But the corral fetch just failed without any indication of why.

[chalcolith]

Massively increased the timeouts on the tests and they ran. @SeanTAllen I think the problem is that the arm build machines are underpowered -- they only have 1gb of memory.

[SeanTAllen]

That seems odd given we are able to build LLVM without much issue on the arm machines.

Perhaps we are accidentally running an amd64 binary that is being emulated. That would take quite a bit of time.

We should verify each of the binaries is arm not amd64. Something isn't right here.

The corral arm windows nightly takes 49 seconds to run.

Is it building libressl that is taking all the time? That only took about 4 to 5 minutes in your PR to add to net_ssl - https://github.com/ponylang/net_ssl/actions/runs/15424265684/job/43407090596?pr=117.

Update:

Yeah all the time is going to "cloning net_ssl" which i think means buildings libressl which only takes 4 to 5 minutes when we did directly in the net_ssl repo:

2025-06-21T15:00:43.6121392Z git cloning github.com/ponylang/net_ssl.git into C:\a\ponyup\ponyup\_repos\github_com_ponylang_net_ssl_git
2025-06-21T16:14:46.8097936Z git checking out @1.3.4 into C:\a\ponyup\ponyup\_corral\github_com_ponylang_net_ssl

that's an hour and 15 minutes. something definitely not right there.

so our PR command in net_ssl is: .\make.ps1 -Command libs 2>&1;
the one from the corral.json for net_ssl is PowerShell.exe -File make.ps1 -Config release -Command libs"

[SeanTAllen]

@chalcolith I did a "test version" of crypto that doesn't build libressl as a post install step. So the only thing as part of that does build libressl is net_ssl. And looks like it was hanging on net_ssl being cloned which might be the building of libressl, but it takes way longer than we see for arm64 windows net_ssl where we only run the powershell command directly in pr.yml for net_ssl.

Its 9 minutes in now. The x64 one was fully done in 5 minutes. I assume this is going to take the full 90 minutes we were seeing, once that is confirm, I will roll back the changes I did to test "only build single LibreSSL".

[SeanTAllen]

@chalcolith and right after i wrote the above, it moved on to other portions of testing. this is going to be much quicker. i can only make the second half of sync on tuesday. perhaps we could chat then (3 pm eastern) or you could make office hours tomorrow as this is looking like it will take considerably more time than amd64 windows but much less than we have been seeing. i will update you on the final time to run if it turns out to be 90 or something ridiculous again despite looking promising.

[SeanTAllen]

22 minutes total when not building libressl as part of the crypto package.

[SeanTAllen]

I think there are one or two things we should do to address the issues that I've indentified...

1. merge crypto and net_ssl into a single library that has 2 packages as part of it. that solves the "we don't want to build more than once" and given both are riding on the same version of a c-library that probably can't easily be used with different versions of it, I think makes sense
2. test out a new feature of corral to control the parallelism for corral operations like clones so that for environments where lots of parallelism/concurrency might make corral operations take longer than serialized, it is available. the test would be to add to corral and then test out here and see if it makes corral cloning operations quicker. if it does, we keep the feature and use it here. if it doesn't, we drop the feature as providing no known value.

[SeanTAllen]

@jemc can you give this a close review to see if there is anything you can see that is missing from this? i just did a review and found one bit that was missing in release.yml and added it.