Dave Abrahams
Dave Abrahams
Ah, with this change: ```diff diff --git a/Package.swift b/Package.swift index 36a041409..e54435a17 100644 --- a/Package.swift +++ b/Package.swift @@ -10,7 +10,7 @@ import PackageDescription /// Settings to be passed to swiftc for...
…and the tests all pass in release build with this change: ```diff diff --git a/Package.swift b/Package.swift index 36a041409..69f77b569 100644 --- a/Package.swift +++ b/Package.swift @@ -10,7 +10,7 @@ import PackageDescription ///...
Turning off WMO has no significant effect on overall testing time.
WMO is not causing a (detected) stack overflow; it's manifesting in an assert. It's possible that there is an undetected stack overflow and UB of course, but that seems very...
Unfortunately, disabling WMO only works _after_ you've built the project once because of https://github.com/apple/swift-package-manager/issues/7274 You can't have it disabled for a clean build.
I've reduced what's needed to reproduce the bug to [this](https://github.com/hylo-lang/hylo/compare/14ebb255d9ad2105bd28291d9e0041f3a2fe42a4...possible-swift-wmo-bug) and reported it [here](https://github.com/apple/swift/issues/70979). It only requires that the standard library ASTs get round-tripped through serialization.
Merging main into this is likely to correct failures
+1, although unless we can think of a collection that _cannot_ be movable when its elements are, that particular example should probably be a trait alias of some kind…
And `swap_at` is not implementable in terms of… well, you can't copy… so… yeah. It might be useful to think for just a moment first about whether there's anything we...
> I think you meant immovable, right? Yeah, that's what I meant. > They are definitely not movable to storage with a longer lifetime, but we could safely move remote...