`type_info` on higher-ranked function pointer produces `TypeId` with strange lifetime.

[theemathas]

In the following code, all the assertions pass. The out variable is set to the TypeId of the return type of for<'a> fn(&'a ()) -> &'a (). I don't know what the lifetime of this reference type is. The type seems to have the following properties:

• It implements RefUnit
• It does not implement IsStatic
• It has a different TypeId from &'static ().

This seems very strange.

#![feature(type_info)]

use std::any::TypeId;
use std::mem::type_info::{Type, TypeKind};

trait RefUnit {}
impl<'a> RefUnit for &'a () {}

trait IsStatic {}
impl<T: 'static> IsStatic for T {}

fn main() {
    let out = const {
        let x = Type::of::<for<'a> fn(&'a ()) -> &'a ()>();
        let TypeKind::FnPtr(fn_ptr) = x.kind else {
            panic!()
        };
        let out: TypeId = fn_ptr.output;
        // out is the TypeId of &'????? ()
        assert!(out.trait_info_of::<dyn RefUnit>().is_some());
        assert!(out.trait_info_of::<dyn IsStatic>().is_none());
        assert!(TypeId::of::<&'static ()>().trait_info_of::<dyn IsStatic>().is_some());
        out
    };
    assert!(out != TypeId::of::<&'static ()>());
}

Meta

Reproducible on the playground with version 1.95.0-nightly (2026-02-24 859951e3c7c9d0322c39)

[oli-obk]

Aaah yea we aren't erasing lifetimes, just skipping binders, so the type ID is indeed of &'^1 ()

[BoxyUwU]

Ideally we should have an assert somewhere that we're not dealing with escaping bound vars to prevent things like this from "working" 🤔