More clear documentation for NonNull<T>

Rephrase and hopefully clarify the discussion of covariance in `NonNull<T>` documentation.
This commit is contained in:
Jonathan Behrens 2020-12-08 17:15:18 -05:00
parent 5e6e1e33a1
commit 0392085f5c

View File

@ -19,12 +19,19 @@ use crate::slice::{self, SliceIndex};
/// as a discriminant -- `Option<NonNull<T>>` has the same size as `*mut T`.
/// However the pointer may still dangle if it isn't dereferenced.
///
/// Unlike `*mut T`, `NonNull<T>` is covariant over `T`. If this is incorrect
/// for your use case, you should include some [`PhantomData`] in your type to
/// provide invariance, such as `PhantomData<Cell<T>>` or `PhantomData<&'a mut T>`.
/// Usually this won't be necessary; covariance is correct for most safe abstractions,
/// such as `Box`, `Rc`, `Arc`, `Vec`, and `LinkedList`. This is the case because they
/// provide a public API that follows the normal shared XOR mutable rules of Rust.
/// Unlike `*mut T`, `NonNull<T>` was chosen to be covariant over `T`. This makes it
/// possible to use `NonNull<T>` when building covariant types, but introduces the
/// risk of unsoundness if used in a type that shouldn't actually be covariant.
/// (The opposite choice was made for `*mut T` even though technically the unsoundness
/// could only be caused by calling unsafe functions.)
///
/// Covariance is correct for most safe abstractions, such as `Box`, `Rc`, `Arc`, `Vec`,
/// and `LinkedList`. This is the case because they provide a public API that follows the
/// normal shared XOR mutable rules of Rust.
///
/// If your type cannot safely be covariant, you must ensure it contains some
/// additional field to provide invariance. Often this field will be a [`PhantomData`]
/// type like `PhantomData<Cell<T>>` or `PhantomData<&'a mut T>`.
///
/// Notice that `NonNull<T>` has a `From` instance for `&T`. However, this does
/// not change the fact that mutating through a (pointer derived from a) shared