Ensure io::Error's bitpacked repr doesn't accidentally impl UnwindSafe

[thomcc]

Sadly, I'm not sure how to easily test that we don't impl a trait, though (or can libstd use where io::Error: !UnwindSafe or something).

Fixes #95203

[rust-highfive]

r? @Mark-Simulacrum

(rust-highfive has picked a reviewer for you, use r? to override)

[thomcc]

Oh, maybe a compile_fail doctest.

[csmoe]

Maybe you can introduce the static_assertion::assert_no_impl_all!(std::io::Error: UnwindSafe) or write the const item yourself like:

 const _: fn() = || {

            trait AmbiguousIfImpl<A> {
                fn some_item() {}
            }

            impl<T: ?Sized> AmbiguousIfImpl<()> for T {}

            impl<T: ?Sized + UnwindSafe> AmbiguousIfImpl<i32> for T {}

            let _ = <io::Error as AmbiguousIfImpl<_>>::some_item;
        };
    };

[thomcc]

Up to the reviewer. I'd rather not add a dependency to fix this if we aren't already depending on it.

As it is, the compile_fail doctest does cover it, I've tested that it ./x.py test library/std fails without the corresponding phantomdata change.

[KamilaBorowska]

I think this could be tested with UI test, see tests in src/test/ui/not-panic directory.

[thomcc]

A downside of using UI tests (compared to the compile_fail doctest) would be that ./x.py test library/std wouldn't test it.

This is not a dealbreaker, but FWIW, I almost never run the full tests, and just let CI get them, but do run the tests for a specific area.

[m-ou-se]

@bors r+

[bors]

📌 Commit 3ac93ab has been approved by m-ou-se

[m-ou-se]

(FWIW: We have compile_fail doctests on private types/fields in more places in core and std, for similar things.)