mikros/rust
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Tweak std::marker docs

Browse Source 
Fixes #29361.
This commit is contained in:
Keegan McAllister 2016-09-11 15:09:05 -07:00
parent dc75933aba
commit b735c1bc78
1 changed files with 266 additions and 153 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

419
src/libcore/marker.rs
View File

@ -8,11 +8,11 @@
// option. This file may not be copied, modified, or distributed
// except according to those terms.
//! Primitive traits and marker types representing basic 'kinds' of types.
//! Primitive traits and types representing basic properties of types.
//!
//! Rust types can be classified in various useful ways according to
//! intrinsic properties of the type. These classifications, often called
//! 'kinds', are represented as traits.
//! their intrinsic properties. These classifications are represented
//! as traits.
#![stable(feature = "rust1", since = "1.0.0")]
@ -22,7 +22,21 @@ use hash::Hasher;
/// Types that can be transferred across thread boundaries.
///
/// This trait is automatically derived when the compiler determines it's appropriate.
/// This trait is automatically implemented when the compiler determines it's
/// appropriate.
///
/// An example of a non-`Send` type is the reference-counting pointer
/// [`rc::Rc`][rc]. If two threads attempt to clone `Rc`s that point to the same
/// reference-counted value, they might try to update the reference count at the
/// same time, which is [undefined behavior][ub] because `Rc` doesn't use atomic
/// operations. Its cousin [`sync::Arc`][arc] does use atomic operations (incurring
/// some overhead) and thus is `Send`.
///
/// See [the Nomicon](../../nomicon/send-and-sync.html) for more details.
///
/// [rc]: ../../std/rc/struct.Rc.html
/// [arc]: ../../std/sync/struct.Arc.html
/// [ub]: ../../reference.html#behavior-considered-undefined
#[stable(feature = "rust1", since = "1.0.0")]
#[lang = "send"]
#[rustc_on_unimplemented = "`{Self}` cannot be sent between threads safely"]
@ -38,10 +52,10 @@ impl<T: ?Sized> !Send for *const T { }
#[stable(feature = "rust1", since = "1.0.0")]
impl<T: ?Sized> !Send for *mut T { }
/// Types with a constant size known at compile-time.
/// Types with a constant size known at compile time.
///
/// All type parameters which can be bounded have an implicit bound of `Sized`. The special syntax
/// `?Sized` can be used to remove this bound if it is not appropriate.
/// All type parameters have an implicit bound of `Sized`. The special syntax
/// `?Sized` can be used to remove this bound if it's not appropriate.
///
/// ```
/// # #![allow(dead_code)]
@ -51,6 +65,26 @@ impl<T: ?Sized> !Send for *mut T { }
/// // struct FooUse(Foo<[i32]>); // error: Sized is not implemented for [i32]
/// struct BarUse(Bar<[i32]>); // OK
/// ```
///
/// The one exception is the implicit `Self` type of a trait, which does not
/// get an implicit `Sized` bound. This is because a `Sized` bound prevents
/// the trait from being used to form a [trait object]:
///
/// ```
/// # #![allow(unused_variables)]
/// trait Foo { }
/// trait Bar: Sized { }
///
/// struct Impl;
/// impl Foo for Impl { }
/// impl Bar for Impl { }
///
/// let x: &Foo = &Impl; // OK
/// // let y: &Bar = &Impl; // error: the trait `Bar` cannot
/// // be made into an object
/// ```
///
/// [trait object]: ../../book/trait-objects.html
#[stable(feature = "rust1", since = "1.0.0")]
#[lang = "sized"]
#[rustc_on_unimplemented = "`{Self}` does not have a constant size known at compile-time"]
@ -59,14 +93,27 @@ pub trait Sized {
// Empty.
}
/// Types that can be "unsized" to a dynamically sized type.
/// Types that can be "unsized" to a dynamically-sized type.
///
/// For example, the sized array type `[i8; 2]` implements `Unsize<[i8]>` and
/// `Unsize<fmt::Debug>`.
///
/// All implementations of `Unsize` are provided automatically by the compiler.
///
/// `Unsize` is used along with [`ops::CoerceUnsized`][coerceunsized] to allow
/// "user-defined" containers such as [`rc::Rc`][rc] to contain dynamically-sized
/// types. See the [DST coercion RFC][RFC982] for more details.
///
/// [coerceunsized]: ../ops/trait.CoerceUnsized.html
/// [rc]: ../../std/rc/struct.Rc.html
/// [RFC982]: https://github.com/rust-lang/rfcs/blob/master/text/0982-dst-coercion.md
#[unstable(feature = "unsize", issue = "27732")]
#[lang="unsize"]
pub trait Unsize<T: ?Sized> {
// Empty.
}
/// Types that can be copied by simply copying bits (i.e. `memcpy`).
/// Types whose values can be duplicated simply by copying bits.
///
/// By default, variable bindings have 'move semantics.' In other
/// words:
@ -87,7 +134,8 @@ pub trait Unsize<T: ?Sized> {
/// However, if a type implements `Copy`, it instead has 'copy semantics':
///
/// ```
/// // we can just derive a `Copy` implementation
/// // We can derive a `Copy` implementation. `Clone` is also required, as it's
/// // a supertrait of `Copy`.
/// #[derive(Debug, Copy, Clone)]
/// struct Foo;
///
@ -100,13 +148,59 @@ pub trait Unsize<T: ?Sized> {
/// println!("{:?}", x); // A-OK!
/// ```
///
/// It's important to note that in these two examples, the only difference is if you are allowed to
/// access `x` after the assignment: a move is also a bitwise copy under the hood.
/// It's important to note that in these two examples, the only difference is whether you
/// are allowed to access `x` after the assignment. Under the hood, both a copy and a move
/// can result in bits being copied in memory, although this is sometimes optimized away.
///
/// ## How can I implement `Copy`?
///
/// There are two ways to implement `Copy` on your type. The simplest is to use `derive`:
///
/// ```
/// #[derive(Copy, Clone)]
/// struct MyStruct;
/// ```
///
/// You can also implement `Copy` and `Clone` manually:
///
/// ```
/// struct MyStruct;
///
/// impl Copy for MyStruct { }
///
/// impl Clone for MyStruct {
/// fn clone(&self) -> MyStruct {
/// *self
/// }
/// }
/// ```
///
/// There is a small difference between the two: the `derive` strategy will also place a `Copy`
/// bound on type parameters, which isn't always desired.
///
/// ## What's the difference between `Copy` and `Clone`?
///
/// Copies happen implicitly, for example as part of an assignment `y = x`. The behavior of
/// `Copy` is not overloadable; it is always a simple bit-wise copy.
///
/// Cloning is an explicit action, `x.clone()`. The implementation of [`Clone`][clone] can
/// provide any type-specific behavior necessary to duplicate values safely. For example,
/// the implementation of `Clone` for [`String`][string] needs to copy the pointed-to string
/// buffer in the heap. A simple bitwise copy of `String` values would merely copy the
/// pointer, leading to a double free down the line. For this reason, `String` is `Clone`
/// but not `Copy`.
///
/// `Clone` is a supertrait of `Copy`, so everything which is `Copy` must also implement
/// `Clone`. If a type is `Copy` then its `Clone` implementation need only return `*self`
/// (see the example above).
///
/// [clone]: ../clone/trait.Clone.html
/// [string]: ../../std/string/struct.String.html
///
/// ## When can my type be `Copy`?
///
/// A type can implement `Copy` if all of its components implement `Copy`. For example, this
/// `struct` can be `Copy`:
/// struct can be `Copy`:
///
/// ```
/// # #[allow(dead_code)]
@ -116,7 +210,8 @@ pub trait Unsize<T: ?Sized> {
/// }
/// ```
///
/// A `struct` can be `Copy`, and `i32` is `Copy`, so therefore, `Point` is eligible to be `Copy`.
/// A struct can be `Copy`, and `i32` is `Copy`, therefore `Point` is eligible to be `Copy`.
/// By contrast, consider
///
/// ```
/// # #![allow(dead_code)]
@ -126,57 +221,35 @@ pub trait Unsize<T: ?Sized> {
/// }
/// ```
///
/// The `PointList` `struct` cannot implement `Copy`, because [`Vec<T>`] is not `Copy`. If we
/// The struct `PointList` cannot implement `Copy`, because [`Vec<T>`] is not `Copy`. If we
/// attempt to derive a `Copy` implementation, we'll get an error:
///
/// ```text
/// the trait `Copy` may not be implemented for this type; field `points` does not implement `Copy`
/// ```
///
/// ## When can my type _not_ be `Copy`?
/// ## When *can't* my type be `Copy`?
///
/// Some types can't be copied safely. For example, copying `&mut T` would create an aliased
/// mutable reference, and copying [`String`] would result in two attempts to free the same buffer.
/// mutable reference. Copying [`String`] would duplicate responsibility for managing the `String`'s
/// buffer, leading to a double free.
///
/// Generalizing the latter case, any type implementing [`Drop`] can't be `Copy`, because it's
/// managing some resource besides its own [`size_of::<T>()`] bytes.
///
/// ## What if I derive `Copy` on a type that can't?
/// If you try to implement `Copy` on a struct or enum containing non-`Copy` data, you will get a
/// compile-time error. Specifically, with structs you'll get [E0204] and with enums you'll get
/// [E0205].
///
/// If you try to derive `Copy` on a struct or enum, you will get a compile-time error.
/// Specifically, with structs you'll get [E0204](https://doc.rust-lang.org/error-index.html#E0204)
/// and with enums you'll get [E0205](https://doc.rust-lang.org/error-index.html#E0205).
/// [E0204]: https://doc.rust-lang.org/error-index.html#E0204
/// [E0205]: https://doc.rust-lang.org/error-index.html#E0205
///
/// ## When should my type be `Copy`?
/// ## When *should* my type be `Copy`?
///
/// Generally speaking, if your type _can_ implement `Copy`, it should. There's one important thing
/// to consider though: if you think your type may _not_ be able to implement `Copy` in the future,
/// then it might be prudent to not implement `Copy`. This is because removing `Copy` is a breaking
/// change: that second example would fail to compile if we made `Foo` non-`Copy`.
///
/// ## Derivable
///
/// This trait can be used with `#[derive]` if all of its components implement `Copy` and the type.
///
/// ## How can I implement `Copy`?
///
/// There are two ways to implement `Copy` on your type:
///
/// ```
/// #[derive(Copy, Clone)]
/// struct MyStruct;
/// ```
///
/// and
///
/// ```
/// struct MyStruct;
/// impl Copy for MyStruct {}
/// impl Clone for MyStruct { fn clone(&self) -> MyStruct { *self } }
/// ```
///
/// There is a small difference between the two: the `derive` strategy will also place a `Copy`
/// bound on type parameters, which isn't always desired.
/// Generally speaking, if your type _can_ implement `Copy`, it should. Keep in mind, though,
/// that implementing `Copy` is part of the public API of your type. If the type might become
/// non-`Copy` in the future, it could be prudent to omit the `Copy` implementation now, to
/// avoid a breaking API change.
///
/// [`Vec<T>`]: ../../std/vec/struct.Vec.html
/// [`String`]: ../../std/string/struct.String.html
@ -188,64 +261,74 @@ pub trait Copy : Clone {
// Empty.
}
/// Types that can be safely shared between threads when aliased.
/// Types for which it is safe to share references between threads.
///
/// This trait is automatically implemented when the compiler determines
/// it's appropriate.
///
/// The precise definition is: a type `T` is `Sync` if `&T` is
/// thread-safe. In other words, there is no possibility of data races
/// when passing `&T` references between threads.
/// [`Send`][send]. In other words, if there is no possibility of
/// [undefined behavior][ub] (including data races) when passing
/// `&T` references between threads.
///
/// As one would expect, primitive types like [`u8`] and [`f64`] are all
/// `Sync`, and so are simple aggregate types containing them (like
/// tuples, structs and enums). More instances of basic `Sync` types
/// include "immutable" types like `&T` and those with simple
/// inherited mutability, such as [`Box<T>`], [`Vec<T>`] and most other
/// collection types. (Generic parameters need to be `Sync` for their
/// container to be `Sync`.)
/// As one would expect, primitive types like [`u8`][u8] and [`f64`][f64]
/// are all `Sync`, and so are simple aggregate types containing them,
/// like tuples, structs and enums. More examples of basic `Sync`
/// types include "immutable" types like `&T`, and those with simple
/// inherited mutability, such as [`Box<T>`][box], [`Vec<T>`][vec] and
/// most other collection types. (Generic parameters need to be `Sync`
/// for their container to be `Sync`.)
///
/// A somewhat surprising consequence of the definition is `&mut T` is
/// `Sync` (if `T` is `Sync`) even though it seems that it might
/// provide unsynchronized mutation. The trick is a mutable reference
/// stored in an aliasable reference (that is, `& &mut T`) becomes
/// read-only, as if it were a `& &T`, hence there is no risk of a data
/// race.
/// A somewhat surprising consequence of the definition is that `&mut T`
/// is `Sync` (if `T` is `Sync`) even though it seems like that might
/// provide unsynchronized mutation. The trick is that a mutable
/// reference behind a shared reference (that is, `& &mut T`)
/// becomes read-only, as if it were a `& &T`. Hence there is no risk
/// of a data race.
///
/// Types that are not `Sync` are those that have "interior
/// mutability" in a non-thread-safe way, such as [`Cell`] and [`RefCell`]
/// in [`std::cell`]. These types allow for mutation of their contents
/// even when in an immutable, aliasable slot, e.g. the contents of
/// [`&Cell<T>`][`Cell`] can be [`.set`], and do not ensure data races are
/// impossible, hence they cannot be `Sync`. A higher level example
/// of a non-`Sync` type is the reference counted pointer
/// [`std::rc::Rc`][`Rc`], because any reference [`&Rc<T>`][`Rc`] can clone a new
/// reference, which modifies the reference counts in a non-atomic
/// way.
/// mutability" in a non-thread-safe form, such as [`cell::Cell`][cell]
/// and [`cell::RefCell`][refcell]. These types allow for mutation of
/// their contents even through an immutable, shared reference. For
/// example the `set` method on `Cell<T>` takes `&self`, so it requires
/// only a shared reference `&Cell<T>`. The method performs no
/// synchronization, thus `Cell` cannot be `Sync`.
///
/// Another example of a non-`Sync` type is the reference-counting
/// pointer [`rc::Rc`][rc]. Given any reference `&Rc<T>`, you can clone
/// a new `Rc<T>`, modifying the reference counts in a non-atomic way.
///
/// For cases when one does need thread-safe interior mutability,
/// types like the atomics in [`std::sync`][`sync`] and [`Mutex`] / [`RwLock`] in
/// the [`sync`] crate do ensure that any mutation cannot cause data
/// races. Hence these types are `Sync`.
/// Rust provides [atomic data types], as well as explicit locking via
/// [`sync::Mutex`][mutex] and [`sync::RWLock`][rwlock]. These types
/// ensure that any mutation cannot cause data races, hence the types
/// are `Sync`. Likewise, [`sync::Arc`][arc] provides a thread-safe
/// analogue of `Rc`.
///
/// Any types with interior mutability must also use the [`std::cell::UnsafeCell`]
/// wrapper around the value(s) which can be mutated when behind a `&`
/// reference; not doing this is undefined behavior (for example,
/// [`transmute`]-ing from `&T` to `&mut T` is invalid).
/// Any types with interior mutability must also use the
/// [`cell::UnsafeCell`][unsafecell] wrapper around the value(s) which
/// can be mutated through a shared reference. Failing to doing this is
/// [undefined behavior][ub]. For example, [`transmute`][transmute]-ing
/// from `&T` to `&mut T` is invalid.
///
/// This trait is automatically derived when the compiler determines it's appropriate.
/// See [the Nomicon](../../nomicon/send-and-sync.html) for more
/// details about `Sync`.
///
/// [`u8`]: ../../std/primitive.u8.html
/// [`f64`]: ../../std/primitive.f64.html
/// [`Vec<T>`]: ../../std/vec/struct.Vec.html
/// [`Box<T>`]: ../../std/boxed/struct.Box.html
/// [`Cell`]: ../../std/cell/struct.Cell.html
/// [`RefCell`]: ../../std/cell/struct.RefCell.html
/// [`std::cell`]: ../../std/cell/index.html
/// [`.set`]: ../../std/cell/struct.Cell.html#method.set
/// [`Rc`]: ../../std/rc/struct.Rc.html
/// [`sync`]: ../../std/sync/index.html
/// [`Mutex`]: ../../std/sync/struct.Mutex.html
/// [`RwLock`]: ../../std/sync/struct.RwLock.html
/// [`std::cell::UnsafeCell`]: ../../std/cell/struct.UnsafeCell.html
/// [`transmute`]: ../../std/mem/fn.transmute.html
/// [send]: trait.Send.html
/// [u8]: ../../std/primitive.u8.html
/// [f64]: ../../std/primitive.f64.html
/// [box]: ../../std/boxed/struct.Box.html
/// [vec]: ../../std/vec/struct.Vec.html
/// [cell]: ../cell/struct.Cell.html
/// [refcell]: ../cell/struct.RefCell.html
/// [rc]: ../../std/rc/struct.Rc.html
/// [arc]: ../../std/sync/struct.Arc.html
/// [atomic data types]: ../sync/atomic/index.html
/// [mutex]: ../../std/sync/struct.Mutex.html
/// [rwlock]: ../../std/sync/struct.RwLock.html
/// [unsafecell]: ../cell/struct.UnsafeCell.html
/// [ub]: ../../reference.html#behavior-considered-undefined
/// [transmute]: ../../std/mem/fn.transmute.html
#[stable(feature = "rust1", since = "1.0.0")]
#[lang = "sync"]
#[rustc_on_unimplemented = "`{Self}` cannot be shared between threads safely"]
@ -314,29 +397,30 @@ macro_rules! impls{
)
}
/// `PhantomData<T>` allows you to describe that a type acts as if it stores a value of type `T`,
/// even though it does not. This allows you to inform the compiler about certain safety properties
/// of your code.
/// Zero-sized type used to mark things that "act like" they own a `T`.
///
/// For a more in-depth explanation of how to use `PhantomData<T>`, please see [the Nomicon].
/// Adding a `PhantomData<T>` field to your type tells the compiler that your
/// type acts as though it stores a value of type `T`, even though it doesn't
/// really. This information is used when computing certain safety properties.
///
/// [the Nomicon]: ../../nomicon/phantom-data.html
/// For a more in-depth explanation of how to use `PhantomData<T>`, please see
/// [the Nomicon](../../nomicon/phantom-data.html).
///
/// # A ghastly note 👻👻👻
///
/// Though they both have scary names, `PhantomData<T>` and 'phantom types' are related, but not
/// identical. Phantom types are a more general concept that don't require `PhantomData<T>` to
/// implement, but `PhantomData<T>` is the most common way to implement them in a correct manner.
/// Though they both have scary names, `PhantomData` and 'phantom types' are
/// related, but not identical. A phantom type parameter is simply a type
/// parameter which is never used. In Rust, this often causes the compiler to
/// complain, and the solution is to add a "dummy" use by way of `PhantomData`.
///
/// # Examples
///
/// ## Unused lifetime parameter
/// ## Unused lifetime parameters
///
/// Perhaps the most common time that `PhantomData` is required is
/// with a struct that has an unused lifetime parameter, typically as
/// part of some unsafe code. For example, here is a struct `Slice`
/// that has two pointers of type `*const T`, presumably pointing into
/// an array somewhere:
/// Perhaps the most common use case for `PhantomData` is a struct that has an
/// unused lifetime parameter, typically as part of some unsafe code. For
/// example, here is a struct `Slice` that has two pointers of type `*const T`,
/// presumably pointing into an array somewhere:
///
/// ```ignore
/// struct Slice<'a, T> {
@ -350,7 +434,7 @@ macro_rules! impls{
/// intent is not expressed in the code, since there are no uses of
/// the lifetime `'a` and hence it is not clear what data it applies
/// to. We can correct this by telling the compiler to act *as if* the
/// `Slice` struct contained a borrowed reference `&'a T`:
/// `Slice` struct contained a reference `&'a T`:
///
/// ```
/// use std::marker::PhantomData;
@ -359,29 +443,53 @@ macro_rules! impls{
/// struct Slice<'a, T: 'a> {
/// start: *const T,
/// end: *const T,
/// phantom: PhantomData<&'a T>
/// phantom: PhantomData<&'a T>,
/// }
/// ```
///
/// This also in turn requires that we annotate `T:'a`, indicating
/// that `T` is a type that can be borrowed for the lifetime `'a`.
/// This also in turn requires the annotation `T: 'a`, indicating
/// that any references in `T` are valid over the lifetime `'a`.
///
/// ## Unused type parameters
///
/// It sometimes happens that there are unused type parameters that
/// indicate what type of data a struct is "tied" to, even though that
/// data is not actually found in the struct itself. Here is an
/// example where this arises when handling external resources over a
/// foreign function interface. `PhantomData<T>` can prevent
/// mismatches by enforcing types in the method implementations:
/// When initializing a `Slice` you simply provide the value
/// `PhantomData` for the field `phantom`:
///
/// ```
/// # #![allow(dead_code)]
/// # trait ResType { fn foo(&self); }
/// # use std::marker::PhantomData;
/// # struct Slice<'a, T: 'a> {
/// # start: *const T,
/// # end: *const T,
/// # phantom: PhantomData<&'a T>,
/// # }
/// fn borrow_vec<'a, T>(vec: &'a Vec<T>) -> Slice<'a, T> {
/// let ptr = vec.as_ptr();
/// Slice {
/// start: ptr,
/// end: unsafe { ptr.offset(vec.len() as isize) },
/// phantom: PhantomData,
/// }
/// }
/// ```
///
/// ## Unused type parameters
///
/// It sometimes happens that you have unused type parameters which
/// indicate what type of data a struct is "tied" to, even though that
/// data is not actually found in the struct itself. Here is an
/// example where this arises with [FFI]. The foreign interface uses
/// handles of type `*mut ()` to refer to Rust values of different
/// types. We track the Rust type using a phantom type parameter on
/// the struct `ExternalResource` which wraps a handle.
///
/// [FFI]: ../../book/ffi.html
///
/// ```
/// # #![allow(dead_code)]
/// # trait ResType { }
/// # struct ParamType;
/// # mod foreign_lib {
/// # pub fn new(_: usize) -> *mut () { 42 as *mut () }
/// # pub fn do_stuff(_: *mut (), _: usize) {}
/// # pub fn new(_: usize) -> *mut () { 42 as *mut () }
/// # pub fn do_stuff(_: *mut (), _: usize) {}
/// # }
/// # fn convert_params(_: ParamType) -> usize { 42 }
/// use std::marker::PhantomData;
@ -408,21 +516,20 @@ macro_rules! impls{
/// }
/// ```
///
/// ## Indicating ownership
/// ## Ownership and the drop check
///
/// Adding a field of type `PhantomData<T>` also indicates that your
/// struct owns data of type `T`. This in turn implies that when your
/// struct is dropped, it may in turn drop one or more instances of
/// the type `T`, though that may not be apparent from the other
/// structure of the type itself. This is commonly necessary if the
/// structure is using a raw pointer like `*mut T` whose referent
/// may be dropped when the type is dropped, as a `*mut T` is
/// otherwise not treated as owned.
/// Adding a field of type `PhantomData<T>` indicates that your
/// type owns data of type `T`. This in turn implies that when your
/// type is dropped, it may drop one or more instances of the type
/// `T`. This has bearing on the Rust compiler's [drop check]
/// analysis.
///
/// If your struct does not in fact *own* the data of type `T`, it is
/// better to use a reference type, like `PhantomData<&'a T>`
/// (ideally) or `PhantomData<*const T>` (if no lifetime applies), so
/// as not to indicate ownership.
///
/// [drop check]: ../../nomicon/dropck.html
#[lang = "phantom_data"]
#[stable(feature = "rust1", since = "1.0.0")]
pub struct PhantomData<T:?Sized>;
@ -438,10 +545,13 @@ mod impls {
/// Types that can be reflected over.
///
/// This trait is implemented for all types. Its purpose is to ensure
/// that when you write a generic function that will employ
/// reflection, that must be reflected (no pun intended) in the
/// generic bounds of that function. Here is an example:
/// By "reflection" we mean use of the [`Any`][any] trait, or related
/// machinery such as [`TypeId`][typeid].
///
/// `Reflect` is implemented for all types. Its purpose is to ensure
/// that when you write a generic function that will employ reflection,
/// that must be reflected (no pun intended) in the generic bounds of
/// that function.
///
/// ```
/// #![feature(reflect_marker)]
@ -455,21 +565,24 @@ mod impls {
/// }
/// ```
///
/// Without the declaration `T: Reflect`, `foo` would not type check
/// (note: as a matter of style, it would be preferable to write
/// `T: Any`, because `T: Any` implies `T: Reflect` and `T: 'static`, but
/// we use `Reflect` here to show how it works). The `Reflect` bound
/// thus serves to alert `foo`'s caller to the fact that `foo` may
/// behave differently depending on whether `T = u32` or not. In
/// particular, thanks to the `Reflect` bound, callers know that a
/// function declared like `fn bar<T>(...)` will always act in
/// precisely the same way no matter what type `T` is supplied,
/// because there are no bounds declared on `T`. (The ability for a
/// caller to reason about what a function may do based solely on what
/// generic bounds are declared is often called the ["parametricity
/// property"][1].)
/// Without the bound `T: Reflect`, `foo` would not typecheck. (As
/// a matter of style, it would be preferable to write `T: Any`,
/// because `T: Any` implies `T: Reflect` and `T: 'static`, but we
/// use `Reflect` here for illustrative purposes.)
///
/// [1]: http://en.wikipedia.org/wiki/Parametricity
/// The `Reflect` bound serves to alert `foo`'s caller to the
/// fact that `foo` may behave differently depending on whether
/// `T` is `u32` or not. The ability for a caller to reason about what
/// a function may do based solely on what generic bounds are declared
/// is often called the "[parametricity property][param]". Despite the
/// use of `Reflect`, Rust lacks true parametricity because a generic
/// function can, at the very least, call [`mem::size_of`][size_of]
/// without employing any trait bounds whatsoever.
///
/// [any]: ../any/trait.Any.html
/// [typeid]: ../any/struct.TypeId.html
/// [param]: http://en.wikipedia.org/wiki/Parametricity
/// [size_of]: ../mem/fn.size_of.html
#[rustc_reflect_like]
#[unstable(feature = "reflect_marker",
reason = "requires RFC and more experience",