CI failure on OpenBSD

[grynspan]

https://github.com/swiftlang/swift-testing/actions/runs/21258606074/job/61180047164#step:6:1

CI crashed. I don't think I have a way to get a core file or other diagnostics from this workflow.

[3405691582]

I noticed this as well. I'll dig into this when able. FWIW, this is SIGABRT which is interesting, rather than anything worse like catching ILL or TERM or something. This is also happening at the end of the test run I think, not at the start, so at least the output is still useful.

Next steps are to try and reproduce locally, which may be difficult...

[grynspan]

I think I can repro this at desk with the latest toolchain you built for me. I think it's terminating testing Test.cancel() specifically.

[3405691582]

After poking around a little further, I actually think there is a problem with the SuspendingClock implementation in Concurrency on this platform. Either that's the case or it's coincidental. If it is then it's going to require a new toolchain and container image rollout. Next step is to get to the bottom of why SuspendingClock isn't working properly for us here.

[grynspan]

I'd bet a dollar @michael-yuji or @etcwilde knows what's up.

[3405691582]

(This may boil down to a regression or something overlooked in Dispatch, even, since I think some of the Concurrency stuff boils down to that.)

[grynspan]

Did you see their FOSDEM talk? There was a problem with the suspending clock on FreeBSD too, and my suspicion is this problem has the same root cause. Watch https://fosdem.org/2026/schedule/event/YTBHYC-swift-on-freebsd/ if you haven't already!

[3405691582]

I briefly skimmed through the subtitles but haven't had a chance to watch fully. :)

The good news is if this is indeed a Dispatch regression, I can pin the revision earlier for the toolchain build.

[3405691582]

I'm fairly confident with the Dispatch changes, #1555, and a new toolchain build, the GHA workflow should work as intended again, assuming that the signal in the workflow is coming from a timeout. I'll leave this open to track the toolchain update and container rollout.