Skip to content

Commit 469c78b

Browse files
fu5haManishearth
authored andcommitted
improve Pin and Pin::new docs
1 parent 6e88279 commit 469c78b

File tree

1 file changed

+67
-23
lines changed

1 file changed

+67
-23
lines changed

library/core/src/pin.rs

Lines changed: 67 additions & 23 deletions
Original file line numberDiff line numberDiff line change
@@ -945,8 +945,13 @@ use crate::{
945945
///
946946
/// In order to pin a value, we wrap a *pointer to that value* (of some type `Ptr`) in a
947947
/// [`Pin<Ptr>`]. [`Pin<Ptr>`] can wrap any pointer type, forming a promise that the **pointee**
948-
/// will not be *moved* or [otherwise invalidated][subtle-details]. Note that it is impossible
949-
/// to create or misuse a [`Pin<Ptr>`] which can violate this promise without using [`unsafe`].
948+
/// will not be *moved* or [otherwise invalidated][subtle-details]. Note that it is
949+
/// impossible to create or misuse a [`Pin<Ptr>`] to violate this promise without using [`unsafe`].
950+
/// If the pointee value's type implements [`Unpin`], we are free to disregard these requirements
951+
/// entirely and can wrap any pointer to that value in [`Pin`] directly via [`Pin::new`].
952+
/// If the pointee value's type does not implement [`Unpin`], then Rust will not let us use the
953+
/// [`Pin::new`] function directly and we'll need to construct a [`Pin`]-wrapped pointer in one of
954+
/// the more specialized manners discussed below.
950955
///
951956
/// We call such a [`Pin`]-wrapped pointer a **pinning pointer,** (or pinning ref, or pinning
952957
/// [`Box`], etc.) because its existince is the thing that is pinning the underlying pointee in
@@ -956,25 +961,57 @@ use crate::{
956961
/// itself, but rather a pointer to that value! A [`Pin<Ptr>`] does not pin the `Ptr` but rather
957962
/// the pointer's ***pointee** value*.
958963
///
959-
/// For the vast majoriy of Rust types, pinning a value of that type will actually have no effect.
960-
/// This is because the vast majority of types implement the [`Unpin`] trait, which entirely opts
961-
/// all values of that type out of pinning-related guarantees. The most common exception
962-
/// to this is the compiler-generated types that implement [`Future`] for the return value
963-
/// of `async fn`s. These compiler-generated [`Future`]s do not implement [`Unpin`] for reasons
964-
/// explained more in the [`pin` module] docs, but suffice it to say they require the guarantees
965-
/// provided by pinning to be implemented soundly.
964+
/// The most common set of types which require pinning related guarantees for soundness are the
965+
/// state machines that implement [`Future`] for the return value of `async fn`s under the
966+
/// hood. These compiler-generated [`Future`]s may contain self-referrential pointers, one of the
967+
/// most common use cases for [`Pin`]. More details on this point are provided in the
968+
/// [`pin` module] docs, but suffice it to say they require the guarantees provided by pinning to
969+
/// be implemented soundly.
966970
///
967-
/// This requirement in the implementation of `async fn`s means that the [`Future`] trait requires
968-
/// any [`Future`] to be pinned in order to call [`poll`] on it. Therefore, when manually polling
969-
/// a future, you will need to pin it first.
971+
/// This requirement from the implementation of `async fn`s means that the [`Future`] trait
972+
/// requires all calls to [`poll`] to use a <code>self: [Pin]\<&mut Self></code> parameter instead
973+
/// of the usual `&mut self`. Therefore, when manually polling a future, you will need to pin it
974+
/// first.
975+
///
976+
/// You may notice that `async fn`-generated [`Future`]s are only a small percentage of all
977+
/// [`Future`]s that exist, yet we had to modify the signature of [`poll`] for all [`Future`]s
978+
/// to accommodate them. This is unfortunate, but there is a way that the language attempts to
979+
/// alleviate the extra friction that this API choice incurs: the [`Unpin`] trait.
980+
///
981+
/// The vast majority of Rust types have no reason to ever care about being pinned. These
982+
/// types implement the [`Unpin`] trait, which entirely opts all values of that type out of
983+
/// pinning-related guarantees. For values of these types, pinning a value by pointing to it with a
984+
/// [`Pin<Ptr>`] will have no actual effect.
985+
///
986+
/// The reason this distinction exists is exactly to allow APIs like [`Future::poll`] to take a
987+
/// [`Pin<Ptr>`] as an argument for all types while only forcing [`Future`] types that actually
988+
/// care about pinning guarantees pay the ergonomics cost. For the majority of [`Future`] types
989+
/// that don't have a reason to care about being pinned and therefore implement [`Unpin`], the
990+
/// <code>[Pin]\<&mut Self></code> will act exactly like a regular `&mut Self`, allowing direct
991+
/// access to the underlying value. Only types that *don't* implement [`Unpin`] will be restricted.
992+
///
993+
/// ### Pinning a value of a type that implements [`Unpin`]
994+
///
995+
/// If the type of the value you need to "pin" implements [`Unpin`], you can trivially wrap any
996+
/// pointer to that value in a [`Pin`] by calling [`Pin::new`].
997+
///
998+
/// ```
999+
/// use std::pin::Pin;
1000+
///
1001+
/// // Create a value of a type that implements `Unpin`
1002+
/// let mut unpin_future = std::future::ready(5);
1003+
///
1004+
/// // Pin it by creating a pinning mutable reference to it (ready to be `poll`ed!)
1005+
/// let my_pinned_unpin_future: Pin<&mut _> = Pin::new(&mut unpin_future);
1006+
/// ```
9701007
///
9711008
/// ### Pinning a value inside a [`Box`]
9721009
///
973-
/// The simplest and most flexible way to pin a value is to put that value inside a [`Box`] and
974-
/// then turn that [`Box`] into a "pinning [`Box`]" by wrapping it in a [`Pin`].
975-
/// You can do both of these in a single step using [`Box::pin`]. Let's see an example of using
976-
/// this flow to pin a [`Future`] returned from calling an `async fn`, a common use case
977-
/// as described above.
1010+
/// The simplest and most flexible way to pin a value that does not implement [`Unpin`] is to put
1011+
/// that value inside a [`Box`] and then turn that [`Box`] into a "pinning [`Box`]" by wrapping it
1012+
/// in a [`Pin`]. You can do both of these in a single step using [`Box::pin`]. Let's see an
1013+
/// example of using this flow to pin a [`Future`] returned from calling an `async fn`, a common
1014+
/// use case as described above.
9781015
///
9791016
/// ```
9801017
/// use std::pin::Pin;
@@ -1018,8 +1055,8 @@ use crate::{
10181055
///
10191056
/// There are some situations where it is desirable or even required (for example, in a `#[no_std]`
10201057
/// context where you don't have access to the standard library or allocation in general) to
1021-
/// pin a value to its location on the stack. Doing so is possible using the [`pin!`] macro. See
1022-
/// its documentation for more.
1058+
/// pin a value which does not implement [`Unpin`] to its location on the stack. Doing so is
1059+
/// possible using the [`pin!`] macro. See its documentation for more.
10231060
///
10241061
/// ## Layout and ABI
10251062
///
@@ -1032,6 +1069,7 @@ use crate::{
10321069
/// [`pin!`]: crate::pin::pin "pin!"
10331070
/// [`Future`]: crate::future::Future "Future"
10341071
/// [`poll`]: crate::future::Future::poll "Future::poll"
1072+
/// [`Future::poll`]: crate::future::Future::poll "Future::poll"
10351073
/// [`pin` module]: self "pin module"
10361074
/// [`Rc`]: ../../std/rc/struct.Rc.html "Rc"
10371075
/// [`Arc`]: ../../std/sync/struct.Arc.html "Arc"
@@ -1137,7 +1175,10 @@ impl<Ptr: Deref<Target: Unpin>> Pin<Ptr> {
11371175
/// use std::pin::Pin;
11381176
///
11391177
/// let mut val: u8 = 5;
1140-
/// // We can pin the value, since it doesn't care about being moved
1178+
///
1179+
/// // Since `val` doesn't care about being moved, we can safely create a "facade" `Pin`
1180+
/// // which will allow `val` to participate in `Pin`-bound apis without checking that
1181+
/// // pinning guarantees are actually upheld.
11411182
/// let mut pinned: Pin<&mut u8> = Pin::new(&mut val);
11421183
/// ```
11431184
#[inline(always)]
@@ -1161,7 +1202,10 @@ impl<Ptr: Deref<Target: Unpin>> Pin<Ptr> {
11611202
///
11621203
/// let mut val: u8 = 5;
11631204
/// let pinned: Pin<&mut u8> = Pin::new(&mut val);
1164-
/// // Unwrap the pin to get a reference to the value
1205+
///
1206+
/// // Unwrap the pin to get the underlying mutable reference to the value. We can do
1207+
/// // this because `val` doesn't care about being moved, so the `Pin` was just
1208+
/// // a "facade" anyway.
11651209
/// let r = Pin::into_inner(pinned);
11661210
/// assert_eq!(*r, 5);
11671211
/// ```
@@ -1317,7 +1361,7 @@ impl<Ptr: Deref> Pin<Ptr> {
13171361
unsafe { Pin::new_unchecked(&*self.pointer) }
13181362
}
13191363

1320-
/// Unwraps this `Pin<Ptr>` returning the underlying pointer.
1364+
/// Unwraps this `Pin<Ptr>`, returning the underlying `Ptr`.
13211365
///
13221366
/// # Safety
13231367
///
@@ -1330,7 +1374,7 @@ impl<Ptr: Deref> Pin<Ptr> {
13301374
///
13311375
/// Note that you must be able to guarantee that the data pointed to by `Ptr`
13321376
/// will be treated as pinned all the way until its `drop` handler is complete!
1333-
///
1377+
///
13341378
/// *For more information, see the [`pin` module docs][self]*
13351379
///
13361380
/// If the underlying data is [`Unpin`], [`Pin::into_inner`] should be used

0 commit comments

Comments
 (0)