Rollup merge of #125696 - workingjubilee:please-dont-say-you-are-lazy, r=Nilstrieb

Explain differences between `{Once,Lazy}{Cell,Lock}` types

The question of "which once-ish cell-ish type should I use?" has been raised multiple times, and is especially important now that we have stabilized the `LazyCell` and `LazyLock` types. The answer for the `Lazy*` types is that you would be better off using them if you want to use what is by far the most common pattern: initialize it with a single nullary function that you would call at every `get_or_init` site. For everything else there's the `Once*` types.

"For everything else" is a somewhat weak motivation, as it only describes by negation. While contrasting them is inevitable, I feel positive motivations are more understandable. For this, I now offer a distinct example that helps explain why `OnceLock` can be useful, despite `LazyLock` existing: you can do some cool stuff with it that `LazyLock` simply can't support due to its mere definition.

The pair of `std::sync::*Lock`s are usable inside a `static`, and can serve roles in async or multithreaded (or asynchronously multithreaded) programs that `*Cell`s cannot. Because of this, they received most of my attention.

Fixes #124696
Fixes #125615

library/core/src/cell.rs

@@ -82,6 +82,20 @@
 //!
 //! The corresponding [`Sync`] version of `OnceCell<T>` is [`OnceLock<T>`].
 //!
+//! ## `LazyCell<T, F>`
+//!
+//! A common pattern with OnceCell is, for a given OnceCell, to use the same function on every
+//! call to [`OnceCell::get_or_init`] with that cell. This is what is offered by [`LazyCell`],
+//! which pairs cells of `T` with functions of `F`, and always calls `F` before it yields `&T`.
+//! This happens implicitly by simply attempting to dereference the LazyCell to get its contents,
+//! so its use is much more transparent with a place which has been initialized by a constant.
+//!
+//! More complicated patterns that don't fit this description can be built on `OnceCell<T>` instead.
+//!
+//! `LazyCell` works by providing an implementation of `impl Deref` that calls the function,
+//! so you can just use it by dereference (e.g. `*lazy_cell` or `lazy_cell.deref()`).
+//!
+//! The corresponding [`Sync`] version of `LazyCell<T, F>` is [`LazyLock<T, F>`].
 //!
 //! # When to choose interior mutability
 //!
@@ -230,6 +244,7 @@
 //! [`RwLock<T>`]: ../../std/sync/struct.RwLock.html
 //! [`Mutex<T>`]: ../../std/sync/struct.Mutex.html
 //! [`OnceLock<T>`]: ../../std/sync/struct.OnceLock.html
+//! [`LazyLock<T, F>`]: ../../std/sync/struct.LazyLock.html
 //! [`Sync`]: ../../std/marker/trait.Sync.html
 //! [`atomic`]: crate::sync::atomic
 

library/std/src/sync/lazy_lock.rs

@@ -29,34 +29,26 @@ union Data<T, F> {
 /// # Examples
 ///
 /// Initialize static variables with `LazyLock`.
-///
 /// ```
-/// use std::collections::HashMap;
-///
 /// use std::sync::LazyLock;
 ///
-/// static HASHMAP: LazyLock<HashMap<i32, String>> = LazyLock::new(|| {
-///     println!("initializing");
-///     let mut m = HashMap::new();
-///     m.insert(13, "Spica".to_string());
-///     m.insert(74, "Hoyten".to_string());
-///     m
+/// // n.b. static items do not call [`Drop`] on program termination, so this won't be deallocated.
+/// // this is fine, as the OS can deallocate the terminated program faster than we can free memory
+/// // but tools like valgrind might report "memory leaks" as it isn't obvious this is intentional.
+/// static DEEP_THOUGHT: LazyLock<String> = LazyLock::new(|| {
+/// # mod another_crate {
+/// #     pub fn great_question() -> String { "42".to_string() }
+/// # }
+///     // M3 Ultra takes about 16 million years in --release config
+///     another_crate::great_question()
 /// });
 ///
-/// fn main() {
-///     println!("ready");
-///     std::thread::spawn(|| {
-///         println!("{:?}", HASHMAP.get(&13));
-///     }).join().unwrap();
-///     println!("{:?}", HASHMAP.get(&74));
-///
-///     // Prints:
-///     //   ready
-///     //   initializing
-///     //   Some("Spica")
-///     //   Some("Hoyten")
-/// }
+/// // The `String` is built, stored in the `LazyLock`, and returned as `&String`.
+/// let _ = &*DEEP_THOUGHT;
+/// // The `String` is retrieved from the `LazyLock` and returned as `&String`.
+/// let _ = &*DEEP_THOUGHT;
 /// ```
+///
 /// Initialize fields with `LazyLock`.
 /// ```
 /// use std::sync::LazyLock;

library/std/src/sync/mod.rs

@@ -136,7 +136,10 @@
 //! - [`Once`]: Used for a thread-safe, one-time global initialization routine
 //!
 //! - [`OnceLock`]: Used for thread-safe, one-time initialization of a
-//!   global variable.
+//!   variable, with potentially different initializers based on the caller.
+//!
+//! - [`LazyLock`]: Used for thread-safe, one-time initialization of a
+//!   variable, using one nullary initializer function provided at creation.
 //!
 //! - [`RwLock`]: Provides a mutual exclusion mechanism which allows
 //!   multiple readers at the same time, while allowing only one

library/std/src/sync/once_lock.rs

@@ -8,47 +8,17 @@ use crate::sync::Once;
 /// A synchronization primitive which can nominally be written to only once.
 ///
 /// This type is a thread-safe [`OnceCell`], and can be used in statics.
+/// In many simple cases, you can use [`LazyLock<T, F>`] instead to get the benefits of this type
+/// with less effort: `LazyLock<T, F>` "looks like" `&T` because it initializes with `F` on deref!
+/// Where OnceLock shines is when LazyLock is too simple to support a given case, as LazyLock
+/// doesn't allow additional inputs to its function after you call [`LazyLock::new(|| ...)`].
 ///
 /// [`OnceCell`]: crate::cell::OnceCell
+/// [`LazyLock<T, F>`]: crate::sync::LazyLock
+/// [`LazyLock::new(|| ...)`]: crate::sync::LazyLock::new
 ///
 /// # Examples
 ///
-/// Using `OnceLock` to store a function’s previously computed value (a.k.a.
-/// ‘lazy static’ or ‘memoizing’):
-///
-/// ```
-/// use std::sync::OnceLock;
-///
-/// struct DeepThought {
-///     answer: String,
-/// }
-///
-/// impl DeepThought {
-/// #   fn great_question() -> String {
-/// #       "42".to_string()
-/// #   }
-/// #
-///     fn new() -> Self {
-///         Self {
-///             // M3 Ultra takes about 16 million years in --release config
-///             answer: Self::great_question(),
-///         }
-///     }
-/// }
-///
-/// fn computation() -> &'static DeepThought {
-///     // n.b. static items do not call [`Drop`] on program termination, so if
-///     // [`DeepThought`] impls Drop, that will not be used for this instance.
-///     static COMPUTATION: OnceLock<DeepThought> = OnceLock::new();
-///     COMPUTATION.get_or_init(|| DeepThought::new())
-/// }
-///
-/// // The `DeepThought` is built, stored in the `OnceLock`, and returned.
-/// let _ = computation().answer;
-/// // The `DeepThought` is retrieved from the `OnceLock` and returned.
-/// let _ = computation().answer;
-/// ```
-///
 /// Writing to a `OnceLock` from a separate thread:
 ///
 /// ```
@@ -73,6 +43,55 @@ use crate::sync::Once;
 ///     Some(&12345),
 /// );
 /// ```
+///
+/// You can use `OnceLock` to implement a type that requires "append-only" logic:
+///
+/// ```
+/// use std::sync::{OnceLock, atomic::{AtomicU32, Ordering}};
+/// use std::thread;
+///
+/// struct OnceList<T> {
+///     data: OnceLock<T>,
+///     next: OnceLock<Box<OnceList<T>>>,
+/// }
+/// impl<T> OnceList<T> {
+///     const fn new() -> OnceList<T> {
+///         OnceList { data: OnceLock::new(), next: OnceLock::new() }
+///     }
+///     fn push(&self, value: T) {
+///         // FIXME: this impl is concise, but is also slow for long lists or many threads.
+///         // as an exercise, consider how you might improve on it while preserving the behavior
+///         if let Err(value) = self.data.set(value) {
+///             let next = self.next.get_or_init(|| Box::new(OnceList::new()));
+///             next.push(value)
+///         };
+///     }
+///     fn contains(&self, example: &T) -> bool
+///     where
+///         T: PartialEq,
+///     {
+///         self.data.get().map(|item| item == example).filter(|v| *v).unwrap_or_else(|| {
+///             self.next.get().map(|next| next.contains(example)).unwrap_or(false)
+///         })
+///     }
+/// }
+///
+/// // Let's exercise this new Sync append-only list by doing a little counting
+/// static LIST: OnceList<u32> = OnceList::new();
+/// static COUNTER: AtomicU32 = AtomicU32::new(0);
+///
+/// let vec = (0..thread::available_parallelism().unwrap().get()).map(|_| thread::spawn(|| {
+///     while let i @ 0..=1000 = COUNTER.fetch_add(1, Ordering::Relaxed) {
+///         LIST.push(i);
+///     }
+/// })).collect::<Vec<thread::JoinHandle<_>>>();
+/// vec.into_iter().for_each(|handle| handle.join().unwrap());
+///
+/// for i in 0..=1000 {
+///     assert!(LIST.contains(&i));
+/// }
+///
+/// ```
 #[stable(feature = "once_cell", since = "1.70.0")]
 pub struct OnceLock<T> {
     once: Once,