sagudev

> > (Although there is still some small race, but I do not think it would be problematic in practice) > > I’m not so sure. The consequence of a...

I am wrong, It could also happen for first decision task to select self-hosted runner, then before is actually dispatched for second decision task to also get self-hosted runner. We...

> But the self-hosted job being cancelled [still makes the whole workflow cancelled](https://github.com/delan/servo/actions/runs/10450947049/job/28936341577) maybe adding result job to windows-auto (that will handle cancellation of self-hostet runner correctly) will cause for...

Maybe we can assign labels to unlabeled runners (this would be uniq_id) in decision task (using job concurrency to only have one job doing labeling at time), and then run...

> Already tried that in [this build](https://github.com/delan/servo/actions/runs/10446746640) ([commit](https://github.com/delan/servo/commit/8c054cac3ddfa22f04a5d1e77dc057c4140f375b)), sadly it didn’t help with the overall result of the workflow. Ah, you are right. And apparently cancellation from `cancel-in-progress` also causes...

> That said, would we still be doing the fallback by setting an output in the decision job, then using that in runs-on? Yes > If so, we still have...

I think secrets should be available in PRs too. See wpt-export for example: https://github.com/servo/servo/blob/8e224cb4d3730899eba2f01c26cb1bd43caa8812/.github/workflows/pull-request-wpt-export.yml#L37 but I will open PR anyway.

I am not able to test this with pushing to try on servo/servo: https://github.com/servo/servo/actions/runs/10507371415/job/29108982963. Are runners and secrets set?

NOTE-TO-SELF: Another interesting thing is this: https://github.com/servo/mozjs/blob/3e8e9d38531e74c75b3ad4921e895c9e8a496082/mozjs/src/gc/root.rs#L194-L196 where we create Handle from MutableHandle. I am not 100% sure how implicit lifetimes works, but I would feel safer if we explicitly...

> Meanwhile in &'a RootedGuard