return the correct type for closures in type_of

[lcnr]

A bit unhappy about the way typeck::check_crate works rn. Would have preferred to not change CollectItemTypesVisitor in this way.

r? @nikomatsakis

[nikomatsakis]

@lcnr This feels like a step backward to me. Can you say a bit more about what's motivating it? In general, I feel like the way things are expected to work is that type_of returns the "generic template" for the type.

[lcnr]

In general, I feel like the way things are expected to work is that type_of returns the "generic template" for the type.

yes, that is also my mental model for type_of. But before this PR, that isn't actually the case.
Before this, type_of returned ty::Closure with substs containing the dummy params for closure_kind, closure_signature, and upvars. To my knowledge, there is not really anything sensible you can do with these substs.

With this change you can now actually call type_of(closure_def_id).subst(tcx, substs) and get the expected result¹.
I am using this in some wip rework of polymorphization where this is needed. If you want, we can talk more about this in sync.

¹: that is a more precise description of my mental model for type_of: I expect subst to result in a correct instance of the type

[nikomatsakis]

@lcnr that makes sense, but I'm a bit confused. Why did that not work before? I guess maybe the question is "what are these substs you are using". I would have expected them to be the closure substs, which would then give a value for those closure type arguments and so forth, right?

[lcnr]

here's an example in case that might help

fn foo<T: Into<i32>, U>(x: T) -> i32 {
    let x = |y: i32| y + x.into();
    x(3)
}

before this PR, the type returned by type_of for |y: i32| y + x.into() was

ty::Closure(def_id, [T, U, <closure_kind>, <closure_signature>, <upvars>])

while with this PR, the type returned by type_of for |y: i32| y + x.into() is

ty::Closure(def_id, [T, U, i32, extern "rust-call" fn((i32,)) -> i32, (T,)])

Before this PR, the only way subst would be correct for the type of x is that we first substitute <closure_kind>, <closure_signature>, <upvars> with their actual values and then , using the result, call subst on the type of x.

With this PR, we can simply call subst with arguments for T and U.

[lcnr]

opened an issue for a more complex fix in #92617

@bors r=nikomatsakis rollup

[bors]

📌 Commit 6bd534d has been approved by nikomatsakis