Change location of where clause on GATs

[jackh726]

Closes #89122

Blocked on lang FCP

r? @nikomatsakis

[bors]

☔ The latest upstream changes (presumably #90188) made this pull request unmergeable. Please resolve the merge conflicts.

[nikomatsakis]

As discussed on our call, let's change this to store a usize "split point" instead.

[lcnr]

One concern I have are generic constants, for example

const SIZE_OF<T>: usize = std::mem::size_of::<T>();

trait IsEqual: 'static {
    const TO<U: 'static>: bool;
}
impl<T: 'static> IsEqual for T {
    const TO<U: 'static>: bool = TypeId::of::<T>() == TypeId::of::<U>();
}

while both of these are currently not allowed in Rust, I personally quite strongly believe that we should add them in the future and want to at least want to keep this option open. Can we unambiguously parse where X after an expression? I would assume yes but am not sure myself.

impl<T: 'static> IsEqual for T {
    const TO<U>: bool = TypeId::of::<T>() == TypeId::of::<U>()
    where
        U: 'static;
}

[nikomatsakis]

@rfcbot fcp cancel

[nikomatsakis]

This shouldn't have been a T-compiler FCP!

[nikomatsakis]

@lcnr

One concern I have are generic constants, for example

const SIZE_OF<T>: usize = std::mem::size_of::<T>();

trait IsEqual: 'static {
    const TO<U: 'static>: bool;
}
impl<T: 'static> IsEqual for T {
    const TO<U: 'static>: bool = TypeId::of::<T>() == TypeId::of::<U>();
}

while both of these are currently not allowed in Rust, I personally quite strongly believe that we should add them in the future and want to at least want to keep this option open. Can we unambiguously parse where X after an expression? I would assume yes but am not sure myself.

Ah, good call! I'm glad you brought that up, we didn't discuss associated constants in the meeting. I'm not worried about parsing ambiguities, the where keyword is a pretty strong signal that we've reached the end of the expression (it's not presently used anywhere else in Rust that I know of).

I'm not 100% sure where I think where should go relative to the = though. It seems to depend a bit on the size of the constant expression.

EDIT: Upon further reflection, I think that I would probably expect where clauses to come after the value for const expressions, too, for the same reasons give above. I don't really expect "complex" const values that span many lines; though I don't doubt they will occur, I suspect it would also be better for folks to factor them out into a const fn.

[petrochenkov]

@nikomatsakis

I'm not 100% sure where I think where should go relative to the = though. It seems to depend a bit on the size of the constant expression.

Do you remember the Centril's proposal to abbreviate fn foo() { result } to fn foo() = result;?
That is a very reasonable thing to do if the body is short, or the body is some kind of match or loop, and I'd still want to have it accepted one day.
(I can't find the relevant RFC PR/issue, unfortunately.)

fn foo() -> u8 = match x {
    A => 1,
    B => 2,
    C => 3,
    _ => 999,
};

What is the preferable position for where in this case?

[petrochenkov]

(Self-assigning, because I have some comments about the implementation as well.)

[camelid]

(I can't find the relevant RFC PR/issue, unfortunately.)

I believe this is it: Centril/rfcs#17

[petrochenkov]

Something is wrong with the FCP here, it's still active and assigned to the compiler team.
@rfcbot fcp cancel

UPD: The previous cancel didn't work because nikomatsakis removed the T-compiler label before writing @rfcbot fcp cancel.

[petrochenkov]

@rfcbot fcp cancel

[rfcbot]

@petrochenkov proposal cancelled.

[bors]

⌛ Testing commit 29f0837501e6f6af372e9274d4ba575409e6f524 with merge 3e5332fe9b563beb405e82b84f4fb648b79b2780...

[bors]

💔 Test failed - checks-actions

[jackh726]

Let's try this again.

@bors r=nikomatsakis

[bors]

📌 Commit d16ec7b has been approved by nikomatsakis

[bors]

⌛ Testing commit d16ec7b with merge ad0d1d7...

[bors]

☀️ Test successful - checks-actions
Approved by: nikomatsakis
Pushing ad0d1d7 to master...

[rust-timer]

Finished benchmarking commit (ad0d1d7): comparison url.

Summary: This benchmark run did not return any relevant results.

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

@rustbot label: -perf-regression