Fix ICE when late-bound lifetimes don't appear in fn signature

[JohnTitor]

PR #135479 changed how late-bound regions are registered for MIR borrowck, switching from for_each_late_bound_region_in_item to iterating over bound_inputs_and_output.bound_vars(). However, for regular functions, the latter only includes late-bound lifetimes that appear in the function's inputs or output, missing unused lifetime parameters.

This caused an ICE when replace_bound_regions_with_nll_infer_vars tried to look up these unregistered regions.

This ensures all late-bound regions are properly registered in the universal regions map before they're needed during region substitution.

Fixes #135845

[rustbot]

This PR changes a file inside tests/crashes. If a crash was fixed, please move into the corresponding ui subdir and add 'Fixes #' to the PR description to autoclose the issue upon merge.

[rustbot]

r? @davidtwco

rustbot has assigned @davidtwco.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rustbot]

⚠️ Warning ⚠️

• There are issue links (such as #123) in the commit messages of the following commits.
  Please move them to the PR description, to avoid spamming the issues with references to the commit, and so this bot can automatically canonicalize them to avoid issues with subtree.
  • 62ab2c2

[davidtwco]

r? types

[jackh726]

Hmm, so. This seems incompatible with the closure_lifetime_binder feature?

I think we should get @lcnr's thoughts on this, on why this switch was made (and if it's appropriate to go back to calling for_each_late_bound_region_in_item on everything).

[lcnr]

However, for regular functions, the latter only includes late-bound lifetimes that appear in the function's inputs or output, missing unused lifetime parameters.

That doesn't seem to be quite true 🤔 from what I can tell all the ICE happen in cases where the fn-sig is malformed. The following compiles:

struct S<'a, T: ?Sized>(&'a T);

fn b<'a>() -> S<'static, u32> {
    S::<'a>(&0)
}

Hmm, so. This seems incompatible with the closure_lifetime_binder feature?

I think we should get @lcnr's thoughts on this, on why this switch was made (and if it's appropriate to go back to calling for_each_late_bound_region_in_item on everything).

I would also expect the same to be true for closures. The Binder should reference all late-bound regions, even if they're unused.

#![feature(closure_lifetime_binder)]
struct S<'a, T: ?Sized>(&'a T);

fn b() {
    for<'a> || -> S<'static, u32> {
        S::<'a>(&0)
    };
}

I personally believe the bug here to be that we don't include late bound regions in the binder if there's an error. We could even add an assert in query fn_sig that all late-bound regions are present in the binder to detect such issues more eagerly

[lcnr]

on why this switch was made

Iirc the main reason behind #135479 was that previously we lazily converted bound regions which feels questionable to me. I switched us to first add nll vars for every late-bound region, and then convert as normal, without having to optionally create new bound regions

[jackh726]

I would also expect the same to be true for closures. The Binder should reference all late-bound regions, even if they're unused.

Okay, yes, that's actually what I would expect too! But wasn't quite sure.

[jackh726]

Okay, so, I think both @lcnr and I are on the same page that this isn't quite the right solution. Going to mark this as waiting on author until an alternate solution is proposed.

@rustbot author