Rollup merge of #46518 - partim:asref-borrow-doc, r=dtolnay

Improve documentation for Borrow

This is the first step in improving the documentation for all the reference conversion traits. It proposes new text for the trait documentation of `Borrow`. Since I feel it is a somewhat radical rewrite and includes a stricter contract for `Borrow` then the previous text—namely that *all* shared traits need to behave the same, not just a select few—, I wanted to get some feedback before continuing.

Apart from the ‘normative’ description, the new text also includes a fairly extensive explanation of how the trait is used in the examples section. I included it because every time I look at how `HashMap` uses the trait, I need to think for a while as the use is a bit twisted. So, I thought having this thinking written down as part of the trait itself might be useful. One could argue that this should go into The Book, and, while I really like having everything important in the docs, I can see the text moved there, too.

So, before I move on: is this new text any good? Do we feel it is correct, useful, comprehensive, and understandable?

(This PR is in response to #44868 and #24140.)
This commit is contained in:
kennytm 2018-03-20 07:15:12 +08:00 committed by GitHub
commit f32f810e6e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -14,24 +14,154 @@
/// A trait for borrowing data.
///
/// In general, there may be several ways to "borrow" a piece of data. The
/// typical ways of borrowing a type `T` are `&T` (a shared borrow) and `&mut T`
/// (a mutable borrow). But types like `Vec<T>` provide additional kinds of
/// borrows: the borrowed slices `&[T]` and `&mut [T]`.
/// In Rust, it is common to provide different representations of a type for
/// different use cases. For instance, storage location and management for a
/// value can be specifically chosen as appropriate for a particular use via
/// pointer types such as [`Box<T>`] or [`Rc<T>`]. Beyond these generic
/// wrappers that can be used with any type, some types provide optional
/// facets providing potentially costly functionality. An example for such a
/// type is [`String`] which adds the ability to extend a string to the basic
/// [`str`]. This requires keeping additional information unnecessary for a
/// simple, immutable string.
///
/// When writing generic code, it is often desirable to abstract over all ways
/// of borrowing data from a given type. That is the role of the `Borrow`
/// trait: if `T: Borrow<U>`, then `&U` can be borrowed from `&T`. A given
/// type can be borrowed as multiple different types. In particular, `Vec<T>:
/// Borrow<Vec<T>>` and `Vec<T>: Borrow<[T]>`.
/// These types provide access to the underlying data through references
/// to the type of that data. They are said to be borrowed as that type.
/// For instance, a [`Box<T>`] can be borrowed as `T` while a [`String`]
/// can be borrowed as `str`.
///
/// If you are implementing `Borrow` and both `Self` and `Borrowed` implement
/// `Hash`, `Eq`, and/or `Ord`, they must produce the same result.
/// Types express that they can be borrowed as some type `T` by implementing
/// `Borrow<T>`, providing a reference to a `T` in the traits
/// [`borrow`] method. A type is free to borrow as several different types.
/// If it wishes to mutably borrow as the type allowing the underlying data
/// to be modified, it can additionally implement [`BorrowMut<T>`].
///
/// `Borrow` is very similar to, but different than, `AsRef`. See
/// [the book][book] for more.
/// Further, when providing implementations for additional traits, it needs
/// to be considered whether they should behave identical to those of the
/// underlying type as a consequence of acting as a representation of that
/// underlying type. Generic code typically uses `Borrow<T>` when it relies
/// on the identical behavior of these additional trait implementations.
/// These traits will likely appear as additional trait bounds.
///
/// [book]: ../../book/first-edition/borrow-and-asref.html
/// If generic code merely needs to work for all types that can
/// provide a reference to related type `T`, it is often better to use
/// [`AsRef<T>`] as more types can safely implement it.
///
/// [`AsRef<T>`]: ../../std/convert/trait.AsRef.html
/// [`BorrowMut<T>`]: trait.BorrowMut.html
/// [`Box<T>`]: ../../std/boxed/struct.Box.html
/// [`Mutex<T>`]: ../../std/sync/struct.Mutex.html
/// [`Rc<T>`]: ../../std/rc/struct.Rc.html
/// [`str`]: ../../std/primitive.str.html
/// [`String`]: ../../std/string/struct.String.html
/// [`borrow`]: #tymethod.borrow
///
/// # Examples
///
/// As a data collection, [`HashMap<K, V>`] owns both keys and values. If
/// the keys actual data is wrapped in a managing type of some kind, it
/// should, however, still be possible to search for a value using a
/// reference to the keys data. For instance, if the key is a string, then
/// it is likely stored with the hash map as a [`String`], while it should
/// be possible to search using a [`&str`][`str`]. Thus, `insert` needs to
/// operate on a `String` while `get` needs to be able to use a `&str`.
///
/// Slightly simplified, the relevant parts of `HashMap<K, V>` look like
/// this:
///
/// ```
/// use std::borrow::Borrow;
/// use std::hash::Hash;
///
/// pub struct HashMap<K, V> {
/// # marker: ::std::marker::PhantomData<(K, V)>,
/// // fields omitted
/// }
///
/// impl<K, V> HashMap<K, V> {
/// pub fn insert(&self, key: K, value: V) -> Option<V>
/// where K: Hash + Eq
/// {
/// # unimplemented!()
/// // ...
/// }
///
/// pub fn get<Q>(&self, k: &Q) -> Option<&V>
/// where
/// K: Borrow<Q>,
/// Q: Hash + Eq + ?Sized
/// {
/// # unimplemented!()
/// // ...
/// }
/// }
/// ```
///
/// The entire hash map is generic over a key type `K`. Because these keys
/// are stored with the hash map, this type has to own the keys data.
/// When inserting a key-value pair, the map is given such a `K` and needs
/// to find the correct hash bucket and check if the key is already present
/// based on that `K`. It therefore requires `K: Hash + Eq`.
///
/// When searching for a value in the map, however, having to provide a
/// reference to a `K` as the key to search for would require to always
/// create such an owned value. For string keys, this would mean a `String`
/// value needs to be created just for the search for cases where only a
/// `str` is available.
///
/// Instead, the `get` method is generic over the type of the underlying key
/// data, called `Q` in the method signature above. It states that `K`
/// borrows as a `Q` by requiring that `K: Borrow<Q>`. By additionally
/// requiring `Q: Hash + Eq`, it signals the requirement that `K` and `Q`
/// have implementations of the `Hash` and `Eq` traits that produce identical
/// results.
///
/// The implementation of `get` relies in particular on identical
/// implementations of `Hash` by determining the keys hash bucket by calling
/// `Hash::hash` on the `Q` value even though it inserted the key based on
/// the hash value calculated from the `K` value.
///
/// As a consequence, the hash map breaks if a `K` wrapping a `Q` value
/// produces a different hash than `Q`. For instance, imagine you have a
/// type that wraps a string but compares ASCII letters ignoring their case:
///
/// ```
/// pub struct CaseInsensitiveString(String);
///
/// impl PartialEq for CaseInsensitiveString {
/// fn eq(&self, other: &Self) -> bool {
/// self.0.eq_ignore_ascii_case(&other.0)
/// }
/// }
///
/// impl Eq for CaseInsensitiveString { }
/// ```
///
/// Because two equal values need to produce the same hash value, the
/// implementation of `Hash` needs to ignore ASCII case, too:
///
/// ```
/// # use std::hash::{Hash, Hasher};
/// # pub struct CaseInsensitiveString(String);
/// impl Hash for CaseInsensitiveString {
/// fn hash<H: Hasher>(&self, state: &mut H) {
/// for c in self.0.as_bytes() {
/// c.to_ascii_lowercase().hash(state)
/// }
/// }
/// }
/// ```
///
/// Can `CaseInsensitiveString` implement `Borrow<str>`? It certainly can
/// provide a reference to a string slice via its contained owned string.
/// But because its `Hash` implementation differs, it behaves differently
/// from `str` and therefore must not, in fact, implement `Borrow<str>`.
/// If it wants to allow others access to the underlying `str`, it can do
/// that via `AsRef<str>` which doesnt carry any extra requirements.
///
/// [`Hash`]: ../../std/hash/trait.Hash.html
/// [`HashMap<K, V>`]: ../../std/collections/struct.HashMap.html
/// [`String`]: ../../std/string/struct.String.html
/// [`str`]: ../../std/primitive.str.html
#[stable(feature = "rust1", since = "1.0.0")]
pub trait Borrow<Borrowed: ?Sized> {
/// Immutably borrows from an owned value.
@ -59,7 +189,11 @@ pub trait Borrow<Borrowed: ?Sized> {
/// A trait for mutably borrowing data.
///
/// Similar to `Borrow`, but for mutable borrows.
/// As a companion to [`Borrow<T>`] this trait allows a type to borrow as
/// an underlying type by providing a mutable reference. See [`Borrow<T>`]
/// for more information on borrowing as another type.
///
/// [`Borrow<T>`]: trait.Borrow.html
#[stable(feature = "rust1", since = "1.0.0")]
pub trait BorrowMut<Borrowed: ?Sized> : Borrow<Borrowed> {
/// Mutably borrows from an owned value.