Type inference falls over on lifetime-generic (HRTB) closures

[cramertj]

This compiles:

trait MyTrait {}
impl<F> MyTrait for F where F: Fn(&i32) -> () {}

fn take_trait<F: MyTrait>(_f: F) {}

fn main() {
    let my_func = |_:&_| {};
    take_trait(my_func);
}

When we take out the &_ type annotation, it breaks:

trait MyTrait {}
impl<F> MyTrait for F where F: Fn(&i32) -> () {}

fn take_trait<F: MyTrait>(_f: F) {}

fn main() {
    let my_func = |_| {};
    take_trait(my_func);
}

It fails to see that the closure should be generic on the lifetime of the input, resulting in this error:

error[E0271]: type mismatch resolving `for<'r> <[closure@<anon>:7:19: 7:25] as std::ops::FnOnce<(&'r i32,)>>::Output == ()`
 --> <anon>:8:5
  |
8 |     take_trait(my_func);
  |     ^^^^^^^^^^ expected bound lifetime parameter , found concrete lifetime
  |
  = note: concrete lifetime that was found is lifetime '_#0r
  = note: required because of the requirements on the impl of `MyTrait` for `[closure@<anon>:7:19: 7:25]`
  = note: required by `take_trait`

error[E0281]: type mismatch: the type `[closure@<anon>:7:19: 7:25]` implements the trait `std::ops::Fn<(_,)>`, but the trait `for<'r> std::ops::Fn<(&'r i32,)>` is required (expected concrete lifetime, found bound lifetime parameter )
 --> <anon>:8:5
  |
8 |     take_trait(my_func);
  |     ^^^^^^^^^^
  |
  = note: required because of the requirements on the impl of `MyTrait` for `[closure@<anon>:7:19: 7:25]`
  = note: required by `take_trait`

error: aborting due to 2 previous errors

The same thing happens when taking lifetime-generic structs as input: the type needs to be manually annotated, or inference will fail. Is there any way to make this work, or to at least get a better error message? I'm interested in helping fix this if someone could point me in the right direction.

[Deewiant]

I just ran into this so I poked around the issue tracker a bit. The inference issue looks like #32600 and #26937 (these three seem like duplicates). #24680 concerns the error message.

[cramertj]

Closing as dup of #32600.