Tracking issue for `const_size_of_val` and `const_align_of_val`

[Gankra]

This is a tracking issue for constness of the following stable API:

// core::mem

pub const fn size_of_val<T: ?Sized>(val: &T) -> usize;
pub const fn align_of_val<T: ?Sized>(val: &T) -> usize;

---

It would be useful for macros to be able to do things like:

static KEY: [u8; size_of_val($byte_string)] = $byte_string;

I believe the rust-objc devs would like something like this to properly set up statics containing messageSend selectors (which are magically processed by the objc runtime by putting them in the right linker sections).

[Gankra]

I believe this is currently blocked on miri becoming the primary/only CTFE provider.

[Gankra]

cc @eddyb @oli-obk

[oli-obk]

I think this is possible without miri const eval, but it would be wasteful to implement in old ctfe

[Gankra]

Yeah I consider "the people who would implement/review this are able but unwilling" to be a blocker

[polarathene]

I believe this is currently blocked on miri becoming the primary/only CTFE provider.

Where can that status be followed? If such a transition were quite a while out, would it be possible to support with only miri and/or feature flag of some kind?

[oli-obk]

Miri is here. So it's mostly a matter of opening a PR by copying over the code from https://github.com/solson/miri/blob/master/miri/intrinsic.rs#L501

[Gankra]

i'm implementing this

[kennytm]

Making these functions const would greatly interfere with custom DST.

Custom DSTs will need the user to implement some custom functions to derive the size. Thus, semantically, making size_of_val const would mean custom DST now need to depend on the postponed rust-lang/rfcs#2337.

Implementation-wise, if we want to support size_of_val for a thin CStr, we would need a const strlen, which means Miri would need to support some selected FFI functions (outside wasm), and also need to perform loop and dereference raw pointers (for wasm target).

Therefore I'm 👎 to making size_of_val and align_of_val const until we have a concrete plan on custom DST. I don't mind having a const fn len() on [T] and str though.

[whitequark]

we would need a const strlen, which means Miri would need to support some selected FFI functions (outside wasm)

Why isn't it possible to implement strlen in Rust instead?

[oli-obk]

Why isn't it possible to implement strlen in Rust instead?

This is done for wasm, but as noted, that would require loops, which have an RFC, but it's not been accepted yet: rust-lang/rfcs#2344 and additionally requires branches, which have an accepted RFC, but no implementation yet

[kennytm]

It also requires dereferencing a raw pointer and offset it, which probably needs the unsafe guideline.

pub unsafe fn strlen(mut s: *const c_char) -> usize {
    let mut n = 0;
    while *s != 0 {      // <-- loop & deref
        n += 1;
        s = s.offset(1); // <-- raw pointer offset
    }
    return n
}

[steveklabnik]

Triage: looks like the original PR didn't get merged. We've been making more and more stuff const fn, what's the current status here?