support anon consts in binders

[lcnr]

closes #72129

The first 2 commits are taken from #79298 which are used in some of the added tests.

This allows us to deal with anonymous constants inside of binders, for example struct Foo<T>(T) where for<'a> [T; { let _: &'a (); 3 }]: Sized;.

What's the problem and how am I trying to solve it?

Looking at the above example, 'a is bound region. As we typeck the anonymous constant separately we would pretty much always work inside of the binder and having bound regions is not supported there.

What we instead want is to treat 'a inside of the anon const as an early bound region. To achieve this we to add some additional generic params to the created AnonConst.

So we pretty much want to desugar the above example into.
Note that we add another generic param to ANON_CONST.

const ANON_CONST<T, 'a> = { let _: &'a (); 3 };
struct Foo<T>(T) where for<'a> [T; ANON_CONST::<T, 'a>]: Sized;

To achieve this we extend hir::AnonConst to remember both it's new synthetic generic parameters as well as the potentially use lifetime arguments originating from a binder. This is now working well enough to warrant a review.

Once the general approach is ironed out I still want to update lifetime resolution to warn for for<'a> [u8; 3 + 4]: Sized that 'a is unused. We also still have to think a bit about perf, though we can probably put the generics into a optional allocation.

r? @matthewjasper

[lcnr]

cc @eddyb

Can we somehow circumvent this? 🤔

I don't think we can just replace the bound vars with placeholders here either.

I think that places which actually try to evaluate the constant do still have to satisfy the ConstEvaluatable predicate but it does allow for

// check-pass
#![feature(const_generics)]
#![allow(incomplete_features)]

trait Baz<const N: usize> {}

fn test<T>() {
    // FIXME(const_generics): This should error.
    let _a: Box<dyn for<'a> Baz<{
        let _: &'a ();
        std::mem::size_of::<T>()
    }>>;
}

fn main() {
    test::<u32>();
}

which is less than ideal

[lcnr]

cc @varkor @eddyb @nikomatsakis, this might also be interesting to you

[bors]

☔ The latest upstream changes (presumably #79511) made this pull request unmergeable. Please resolve the merge conflicts.

Note that reviewers usually do not review pull requests until merge conflicts are resolved! Once you resolve the conflicts, you should change the labels applied by bors to indicate that your PR is ready for review. Post this as a comment to change the labels:

@rustbot modify labels: +S-waiting-on-review -S-waiting-on-author

[cjgillot]

Would replacing AnonConst by &'hir AnonConst<'hir> help with the size?

[lcnr]

yeah, probably 👍

I want to wait for a first review before looking into perf though, so I don't waste time on an approach we don't want in the end

[bors]

☔ The latest upstream changes (presumably #80163) made this pull request unmergeable. Please resolve the merge conflicts.

Note that reviewers usually do not review pull requests until merge conflicts are resolved! Once you resolve the conflicts, you should change the labels applied by bors to indicate that your PR is ready for review. Post this as a comment to change the labels:

@rustbot modify labels: +S-waiting-on-review -S-waiting-on-author

[JohnCSimon]

@rustbot label: -S-waiting-on-review +S-waiting-on-author

[Dylan-DPC-zz]

@lcnr any updates on this?

[lcnr]

want to talk about this with project-const-generic in the near future before doing much here.

[lcnr]

probably want to wait with this pr until @rust-lang/project-const-generics figures out how to deal with unused params in general as that will influence the way we want to implement this change.