dtors_in_dtors_in_dtors test fails on 32bit Windows targets

[RalfJung]

Most Windows targets (both MSVC and GNU) deadlock on this test when running in Miri. Only x86_64-pc-windows-gnu target works fine.

I seem to recall that on the failing targets we do not use "native" thread-locals due to some linker issues or so.

But the real question is, why don't the destructors work the way they should?

[RalfJung]

Oh, something is even stranger. The test works fine when I extract it into a separate Miri test, but fails on the std test suite.

[RalfJung]

Turns out to be a realstd issue. Fixed by rust-lang/rust#106638.