Tracking Issue for raw-pointer-to-reference conversion methods

[RalfJung]

Feature gate: #![feature(as_ref_unchecked)] (or similar)

This is a tracking issue for raw-pointer-to-reference conversion methods (rust-lang/libs-team#342).

With #106116, we have methods for almost all casts/conversions one wants to do on references and pointers, to avoid as casts and prefix operators. Just one direction is missing: turning raw pointers into references. Here we have as_ref/as_mut, but they behave different from &*ptr/&mut *ptr: they return an Option and perform a null-check. (These are the only methods on raw pointers that perform a null check.)

Public API

impl<T> *const T {
  unsafe const fn as_ref_unchecked<'a>(self) -> &'a T {
    &*self
  }
}
impl<T> *mut T {
  unsafe const fn as_ref_unchecked<'a>(self) -> &'a T {
    &*self
  }
  unsafe const fn as_mut_unchecked<'a>(self) -> &'a mut T {
    &mut *self
  }
}

Motivating examples or use cases

Some random examples from a quick grep of the rustc sources:

&mut *(out as *mut &mut dyn PrintBackendInfo)
&mut *(state as *mut &mut dyn FnMut(&[u8]) -> io::Result<()>)
&mut *(self.0 as *mut _)
&mut *(vec as *mut Vec<Library>)
&mut *(value as *mut T as *mut UnsafeCell<T>)
&mut *(s as *mut T).cast::<[T; 1]>()

The examples above then become

out.cast::<&mut dyn PrintBackendInfo>().as_mut_unchecked()
state.cast::<&mut dyn FnMut(&[u8]) -> io::Result<()>>().as_mut_unchecked()
self.0.cast().as_mut_unchecked()
vec.cast::<Vec<Library>>().as_mut_unchecked()
ptr::from_mut(value).cast::<UnsafeCell<T>>().as_mut_unchecked()
ptr::from_mut(s).cast::<[T; 1]>().as_mut_unchecked()

Unfortunately, since the as_mut name is already taken, these are all longer than the prefix variants. But they can be read left-to-right which is an advantage.

Steps / History

• Implementation: Implement ptr_as_ref_unchecked #122492
• Final comment period (FCP)¹
• Stabilization PR

Unresolved Questions

• Is there a shorter name we could use?

Footnotes

1. https://std-dev-guide.rust-lang.org/feature-lifecycle/stabilization.html ↩

[Qix-]

/bikeshed

Is there a shorter name we could use?

I personally think this is fine. Follows the conventions, and most of the cases where I've reached for these methods, they've been inside macros etc. where it matters way less.

I know it's not really an option now (maybe it is?) but personally would have preferred unsafe fn as_ref() do what as_ref_unchecked() does with a try_as_ref() returning the Option<&T> (what it does now). But I wouldn't think that's possible given that it'd break a bunch of people (maybe it could be done in a new edition but not sure it's worth the effort).

[ibraheemdev]

Were deref and deref_mut ever considered? I can see how this could be confusing (and maybe ambiguous?), but these methods are the deref operations, just unsafe.

[RalfJung]

This doesn't deref the pointer though, so that name seems confusing to me.

The Deref trait has &Target return type just because we don't have a type that would represent a "place of type T", but the way it gets invoked by the compiler makes this still behave largely like a place. That cannot be said about a method we call directly manually.

[ibraheemdev]

Sure but it is equivalent to calling Deref::deref on any other pointer type, just unsafe. Similarly * returns a place to T, but is unsafe, so it is consistent in that regard. You generally do not call deref manually but it can be useful for method chaining as a postfix operation (taking advantage of auto-deref).