use correct feature flag for impl-block-level trait bounds on const fn

[RalfJung]

I am not sure what that special hack was needed for, but it doesn't seem needed any more...

This removes the last use of the const_fn feature flag -- Cc #84510
r? @oli-obk

[oli-obk]

Maybe we don't have tests for the situation the comment was referring to? Check if we have a test where a generic impl const block can use its trait bounds in a function, if not, please add one

[RalfJung]

I don't understand what that situation would look like, could you sketch an example?

[oli-obk]

On phone, but i'll try:

impl<T: Foo> const Bar for () {
    fn bar() {
        T::foo()
    } 
}

Declare the Foo and Bar traits appropriately.

[RalfJung]

I did that, but I am not sure if that is the test the comment was referring to... the comment sounded more like a situation where a const Trait bound somewhere was incorrectly not recognized as const and hence, I assume, leads to an error -- IOW, the comment referred to some code incorrectly failing to build (not some code incorrectly building).

impl<T: const Plus> const Plus for Num<T> does not even parse, though.
This code does parse and, to my surprise, fails to build:

#![feature(const_fn_trait_bounds)]
#![feature(const_trait_impl)]
#![feature(const_trait_bound_opt_out)]
#![allow(incomplete_features)]

mod num {
    pub trait Plus {
        fn plus(self, rhs: Self) -> Self;
    }

    struct Num<T>(T);
    impl<T: ?const Plus + Copy> const Plus for Num<T> {
        fn plus(self, rhs: Self) -> Self {
            Num(self.0.plus(rhs.0))
        }
    }
}

However, the error has nothing to do with feature gating:

error[E0015]: calls in constant functions are limited to constant functions, tuple structs and tuple variants
  --> /home/r/src/rust/rustc.2/src/test/ui/rfc-2632-const-trait-impl/call-generic-method-nonconst-opt-out.rs:30:17
   |
LL |             Num(self.0.plus(rhs.0))
   |                 ^^^^^^^^^^^^^^^^^^

error: aborting due to previous error

So, I don't think what this PR does makes things any worse here.

[RalfJung]

I think what is left here is to fix lock_api -- I'll submit a PR there tomorrow (once #84578 is in nightly).

[oli-obk]

This code does parse and, to my surprise, fails to build:

Why are you surprised it fails to build? It should fail, you are using ?const to opt out of constness after all

[oli-obk]

The problem is that it also fails to compile without ?const

[RalfJung]

Why are you surprised it fails to build? It should fail, you are using ?const to opt out of constness after all

Ah, I forgot that things are implicitly const.^^

[RalfJung]

The problem is that it also fails to compile without ?const

Okay, I guess I should remove my new test then? But it says nothing about a feature gate there, so I don't think my PR changes the situation here... it might mean some people need to add feature(const_fn_trait_bounds) but that does not seem unreasonable?

[oli-obk]

The problem is that it also fails to compile without ?const

Okay, I guess I should remove my new test then? But it says nothing about a feature gate there, so I don't think my PR changes the situation here... it might mean some people need to add feature(const_fn_trait_bounds) but that does not seem unreasonable?

we already have a test for this behaviour and possibly even an issue, idk where though

[RalfJung]

we already have a test for this behaviour and possibly even an issue, idk where though

Okay so I removed the new test again. Should I add anything else instead?

[oli-obk]

we already have a test for this behaviour and possibly even an issue, idk where though

Okay so I removed the new test again. Should I add anything else instead?

idk, I don't have the time to really dig into this, so let's just merge as is. It preserves the status quo, so all is good

[RalfJung]

Cc Amanieu/parking_lot#281

[oli-obk]

@bors r+

[bors]

📌 Commit 3c4c5eb has been approved by oli-obk

[bors]

⌛ Testing commit 3c4c5eb with merge 18587b1...

[bors]

☀️ Test successful - checks-actions
Approved by: oli-obk
Pushing 18587b1 to master...