variant of PR 87770 with limited scope for backport.

[pnkfelix]

This version does not attempt to resolve the problem that is illustrated by src/test/ui/const-generics/generic_const_exprs/drop_impl.rs (a problem that relies on nightly features to demonstrate anyway).

My thinking is that this is a better choice for backporting to stable and beta.

Fix #90838 (which was closed by accident by #90840, which only landed on nightly, so far).

[rust-highfive]

r? @Mark-Simulacrum

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

[rust-highfive]

⚠️ Warning ⚠️

• Pull requests are usually filed against the master branch for this repo, but this one is against beta. Please double check that you specified the right target!

[pnkfelix]

(The compiler team itself hasn't discussed whether we would be better off backporting this, or backporting PR #90840 itself. I believe we will resolve that question on Wednesday.)

[Mark-Simulacrum]

I'm probably not a good reviewer here -- r? @jackh726 perhaps?

[jackh726]

This implementation looks good to me, but I don't know if its strictly better than #90840 for backport.

Does the new test you added also pass on master? We probably want that test added regardless.

On one hand, this is somewhat of a revert to #87770, which is "safer". On the other, it's pretty clear that the cause of the issue was just a failure to relate the lifetimes.

I'll leave the decision to backport this or #90840 to the compiler team, but I personally prefer the latter out of consistency. However, r=me on this implementation.

[pnkfelix]

This implementation looks good to me, but I don't know if its strictly better than #90840 for backport.

Does the new test you added also pass on master? We probably want that test added regardless.

The test is taken from #90840. So it already is on master.

[pnkfelix]

(Closing; We'll backport #90840 instead of this.)