downcast beta regression

[est31]

The tests of the downcast crate version 0.9.0 fail on latest beta 1.21, while they work on stable. The error output (full log):

error[E0310]: the parameter type `X` may not live long enough
  --> examples/with_params.rs:11:1
   |
11 | downcast!(<X> Animal<X> where X: Debug);
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   |
   = help: consider adding an explicit lifetime bound `X: 'static`...
note: ...so that the type `Animal<X> + 'static` will meet its required lifetime bounds
  --> examples/with_params.rs:11:1
   |
11 | downcast!(<X> Animal<X> where X: Debug);
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   = note: this error originates in a macro outside of the current crate

cc @fkoep

[est31]

Using bisect-rust I've bisected the regression to 426711d (#43786). @scalexm @nikomatsakis is this an expected regression?

[scalexm]

@fkoep @est31 This is expected, the type parameter X indeed needs a 'static bound so that the trait object type Animal<X> is 'static (required by the StdAny trait). Before #43786, the bound was not checked.

However this code snippet revealed another compiler bug, which may alter the solution I'm proposing (adding a X: 'static where clause). Waiting for a confirmation by @nikomatsakis before we can close this issue.

[nikomatsakis]

So, this change is indeed correcting a soundness violation (#43784), and the change doesn't appear to have caused widespread breakage. Therefore, we are going to leave the code as is. Tagging as relnotes.

---

@scalexm

However this code snippet revealed another compiler bug, which may alter the solution I'm proposing

I guess that you are referring to #44454 -- but I think the solution you are proposing is probably still correct, right?

[scalexm]

@nikomatsakis yes, just forgot to get back on this.