Make [T]::len and str::len const fn

[oli-obk]

r? @gankro

[oli-obk]

cc @Centril

[leonardo-m]

A simple common use case:

fn main() {
    const A: [u32; 3] = [1, 2, 3];
    const B: [u32; A.len()] = [10, 20, 30];
}

[Centril]

Nice!

[Gankra]

Seems solid. I don’t have review rights tho

[oli-obk]

r? @Centril

[oli-obk]

Hot potato passing until success

r? @SimonSapin

[Gankra]

Can you make this assert some equality?

[SimonSapin]

@bors r=SimonSapin,Gankro

[bors]

📌 Commit 2788f66 has been approved by SimonSapin,Gankro

[leonardo-m]

It doesn't work in this equally common case:

#![feature(const_slice_len)]
#![allow(unused_variables)]

fn main() {
    let a: [u8; 3] = [10, 20, 30];
    let b: [u8; a.len()] = Default::default();
    // error[E0435]: attempt to use a non-constant value in a constant
}

[oli-obk]

That will never work with const eval as it is setup currently. Supporting such cases (I'd call them "lazy const eval") would technically be possible but definitely requires an RFC.

[fbstj]

can/should the union be moved outside the unsafe block? it's definition isn't unsafe iirc

[oli-obk]

While that could certainly be done, there's no difference here and this seems very self-contained to me. YMMV, so feel free to open a PR about that if you want :)

[rodrimati1992]

Is this likely to stabilize soon?I need either this or the layout of &[T] and str to be guaranteed,to be able to to construct ffi-safe slices in constants without using hacks.

The hack I am currently using is(approximately):

#[repr(C)]
pub struct StaticSlice<T: 'static> {
    s: &'static [T],
    conversion: extern fn(*const [usize;2])->RSlice<'static,T>,
}

#[repr(C)]
pub struct RSlice<'a,T>{
    ptr:*const T,
    len:usize,
    _marker:PhantomData<&'a T>,
}

Along with a static assertions that &[T] is:

• 2*usize in size
• usize in alignment

The function pointer takes in a *const &[T] casted to *const [usize;2] which the static assertion guarantees is valid.

[oli-obk]

Stabilization is blocked on #51909