rfcs/text/2306-convert-id.md

8.1 KiB

Summary

Adds an identity function pub const fn identity<T>(x: T) -> T { x } as core::convert::identity. The function is also re-exported to std::convert::identity.

Motivation

The identity function is useful

While it might seem strange to have a function that just returns back the input, there are some cases where the function is useful.

Using identity to do nothing among a collection of mappers

When you have collections such as maps or arrays of mapping functions like below and you watch to dispatch to those you sometimes need the identity function as a way of not transforming the input. You can use the identity function to achieve this.

// Let's assume that this and other functions do something non-trivial.
fn do_interesting_stuff(x: u32) -> u32 { .. }

// A dispatch-map of mapping functions:
let mut map = HashMap::new();
map.insert("foo", do_interesting_stuff);
map.insert("bar", other_stuff);
map.insert("baz", identity);

Using identity as a no-op function in a conditional

This reasoning also applies to simpler yes/no dispatch as below:

let mapper = if condition { some_manipulation } else { identity };

// do more interesting stuff inbetween..

do_stuff(42);

Using identity to concatenate an iterator of iterators

We can use the identity function to concatenate an iterator of iterators into a single iterator.

let vec_vec = vec![vec![1, 3, 4], vec![5, 6]];
let iter_iter = vec_vec.into_iter().map(Vec::into_iter);
let concatenated = iter_iter.flat_map(identity).collect::<Vec<_>>();
assert_eq!(vec![1, 3, 4, 5, 6], concatenated);

While the standard library has recently added Iterator::flatten, which you should use instead, to achieve the same semantics, similar situations are likely in the wild and the identity function can be used in those cases.

Using identity to keep the Some variants of an iterator of Option<T>

We can keep all the maybe variants by simply iter.filter_map(identity).

let iter = vec![Some(1), None, Some(3)].into_iter();
let filtered = iter.filter_map(identity).collect::<Vec<_>>();
assert_eq!(vec![1, 3], filtered);

To be clear that you intended to use an identity conversion

If you instead use a closure as in |x| x when you need an identity conversion, it is less clear that this was intentional. With identity, this intent becomes clearer.

The drop function as a precedent

The drop function in core::mem is defined as pub fn drop<T>(_x: T) { }. The same effect can be achieved by writing { _x; }. This presents us with a precedent that such trivial functions are considered useful and includable inside the standard library even though they can be written easily inside a user's crate.

Avoiding repetition in user crates

Here are a few examples of the identity function being defined and used:

There's a smattering of more examples. To reduce duplication, it should be provided in the standard library as a common place it is defined.

Precedent from other languages

There are other languages that include an identity function in their standard libraries, among these are:

  • Haskell, which also exports this to the prelude.
  • Scala, which also exports this to the prelude.
  • Java, which is a widely used language.
  • Idris, which also exports this to the prelude.
  • Ruby, which exports it to what amounts to the top type.
  • Racket
  • Julia
  • R
  • F#
  • Clojure
  • Agda
  • Elm

Guide-level explanation

An identity function is a mapping of one type onto itself such that the output is the same as the input. In other words, a function identity : T -> T for some type T defined as identity(x) = x. This RFC adds such a function for all Sized types in Rust into libcore at the module core::convert and defines it as:

pub const fn identity<T>(x: T) -> T { x }

This function is also re-exported to std::convert::identity.

It is important to note that the input x passed to the function is moved since Rust uses move semantics by default.

Reference-level explanation

An identity function defined as pub const fn identity<T>(x: T) -> T { x } exists as core::convert::identity. The function is also re-exported as std::convert::identity-

Note that the identity function is not always equivalent to a closure such as |x| x since the closure may coerce x into a different type while the identity function never changes the type.

Drawbacks

It is already possible to do this in user code by:

  • using an identity closure: |x| x.
  • writing the identity function as defined in the RFC yourself.

These are contrasted with the motivation for including the function in the standard library.

Rationale and alternatives

The rationale for including this in convert and not mem is that the former generally deals with conversions, and identity conversion" is a used phrase. Meanwhile, mem does not relate to identity other than that both deal with move semantics. Therefore, convert is the better choice. Including it in mem is still an alternative, but as explained, it isn't fitting.

Naming the function id instead of identity is a possibility. This name is however ambiguous with "identifier" and less clear wherefore identifier was opted for.

Unresolved questions

There are no unresolved questions.

Possible future work

A previous iteration of this RFC proposed that the identity function should be added to prelude of both libcore and libstd. However, the library team decided that for the time being, it was not sold on this inclusion. As we gain usage experience with using this function, it is possible to revisit this in the future if the team chances its mind.

The section below details, for posterity, the argument for inclusion that was previously in the motivation.

The case for inclusion in the prelude

Let's compare the effort required, assuming that each letter typed has a uniform cost with respect to effort.

use std::convert::identity; iter.filter_map(identity)

fn identity<T>(x: T) -> T { x } iter.filter_map(identity)

iter.filter_map(::std::convert::identity)

iter.filter_map(identity)

Comparing the length of these lines, we see that there's not much difference in length when defining the function yourself or when importing or using an absolute path. But the prelude-using variant is considerably shorter. To encourage the use of the function, exporting to the prelude is therefore a good idea.

In addition, there's an argument to be made from similarity to other things in core::convert as well as drop all of which are in the prelude. This is especially relevant in the case of drop which is also a trivial function.