RFC: Implementable trait aliases

Author: Jules-Bertholet

text/3437-implementable-trait-alias.md

@@ -0,0 +1,319 @@
+- Feature Name: `trait_alias_impl`
+- Start Date: 2023-05-24
+- RFC PR: [rust-lang/rfcs#3437](https://github.com/rust-lang/rfcs/pull/3437)
+- Rust Issue: [rust-lang/rust#0000](https://github.com/rust-lang/rust/issues/0000)
+
+# Summary
+
+Extend `#![feature(trait_alias)]` to permit `impl` blocks for trait aliases with a single primary trait.
+
+# Motivation
+
+Often, one desires to have a "weak" version of a trait, as well as a "strong" one providing additional guarantees.
+Subtrait relationships are commonly used for this, but they sometimes fall short—expecially when the "strong" version is
+expected to see more use, or was stabilized first.
+
+## Example: AFIT `Send` bound aliases
+
+### With subtraits
+
+Imagine a library, `frob-lib`, that provides a trait with an async method. (Think `tower::Service`.)
+
+```rust
+//! crate frob-lib
+pub trait Frobber {
+    async fn frob(&self);
+}
+```
+
+Most of `frob-lib`'s users will need `Frobber::frob`'s return type to be `Send`,
+so the library wants to make this common case as painless as possible. But non-`Send`
+usage should be supported as well.
+
+`frob-lib`, following the recommended practice, decides to design its API in the following way:
+
+```rust
+//! crate frob-lib
+
+pub trait LocalFrobber {
+    async fn frob(&self);
+}
+
+// or whatever RTN syntax is decided on
+pub trait Frobber: LocalFrobber<frob(..): Send> + Send {}
+impl<T: ?Sized> Frobber for T where T: LocalFrobber<frob(..): Send> + Send {}
+```
+
+These two traits are, in a sense, one trait with two forms: the "weak" `LocalFrobber`, and "strong" `Frobber` that offers an additional `Send` guarantee.
+
+Because `Frobber` (with `Send` bound) is the common case, `frob-lib`'s documentation and examples put it front and center. So naturally, Joe User tries to implement `Frobber` for his own type.
+
+```rust
+//! crate joes-crate
+use frob_lib::Frobber;
+
+struct MyType;
+
+impl Frobber for MyType {
+    async fn frob(&self) {
+        println!("Sloo is 120% klutzed. Initiating brop sequence...")
+    }
+}
+```
+
+But one `cargo check` later, Joe is greeted with:
+
+```
+error[E0277]: the trait bound `MyType: LocalFrobber` is not satisfied
+  --> src/lib.rs:6:18
+   |
+6  | impl Frobber for MyType {
+   |                  ^^^^^^ the trait `LocalFrobber` is not implemented for `MyType`
+
+error[E0407]: method `frob` is not a member of trait `Frobber`
+  --> src/lib.rs:7:5
+   |
+7  | /     async fn frob(&self) {
+8  | |         println!("Sloo is 120% klutzed. Initiating brop sequence...")
+9  | |     }
+   | |_____^ not a member of trait `Frobber`
+```
+
+Joe is confused. "What's a `LocalFrobber`? Isn't that only for non-`Send` use cases? Why do I need to care about all that?" But he eventually figures it out:
+
+```rust
+//! crate joes-crate
+use frob_lib::LocalFrobber;
+
+struct MyType;
+
+impl LocalFrobber for MyType {
+    #[refine]
+    async fn frob(&self) {
+        println!("Sloo is 120% klutzed. Initiating brop sequence...")
+    }
+}
+```
+
+This is distinctly worse. Joe now has to reference both `Frobber` and `LocalFrobber` in his code,
+and (assuming that the final AFIT feature ends up requiring it) also has to write `#[refine]`.
+
+### With today's `#![feature(trait_alias)]`
+
+What if `frob-lib` looked like this instead?
+
+```rust
+//! crate frob-lib
+#![feature(trait_alias)]
+
+pub trait LocalFrobber {
+    async fn frob(&self);
+}
+
+pub trait Frobber = LocalFrobber<frob(..): Send> + Send;
+```
+
+With today's `trait_alias`, it wouldn't make much difference for Joe. He would just get a slightly different
+error message:
+
+```
+error[E0404]: expected trait, found trait alias `Frobber`
+  --> src/lib.rs:6:6
+   |
+6  | impl Frobber for MyType {
+   |      ^^^^^^^ not a trait
+```
+
+## Speculative example: GATification of `Iterator`
+
+*This example relies on some language features that are currently pure speculation.
+Implementable trait aliases are potentially necessary to support this use-case, but not sufficent.*
+
+Ever since the GAT MVP was stabilized, there has been discussion about how to add `LendingIterator` to the standard library, without breaking existing uses of `Iterator`.
+The relationship between `LendingIterator` and `Iterator` is "weak"/"strong"—an `Iterator` is a `LendingIterator` with some extra guarantees about the `Item` associated type.
+
+Now, let's imagine that Rust had some form of "variance bounds",
+that allowed restricting the way in which a type's GAT can depend on said GAT's generic parameters.
+One could then define `Iterator` in terms of `LendingIterator`, like so:
+
+```rust
+//! core::iter
+pub trait LendingIterator {
+    type Item<'a>
+    where
+        Self: 'a;
+
+    fn next(&'a mut self) -> Self::Item<'a>;
+}
+
+pub trait Iterator = LendingIterator
+where
+    // speculative syntax, just for the sake of this example
+    for<'a> Self::Item<'a>: bivariant_in<'a>;
+```
+
+But, as with the previous example, we are foiled by the fact that trait aliases aren't `impl`ementable,
+so this change would break every `impl Iterator` block in existence.
+
+## Speculative example: `Async` trait
+
+There has been some discussion about a variant of the `Future` trait with an `unsafe` poll method, to support structured concurrency ([here](https://rust-lang.github.io/wg-async/vision/roadmap/scopes/capability/variant_async_trait.html) for example). *If* such a change ever happens, then the same "weak"/"strong" relationship will arise: the safe-to-poll `Future` trait would be a "strong" version of the unsafe-to-poll `Async`. As the linked design notes explain, there are major problems with expressing this relationship in today's Rust.
+
+# Guide-level explanation
+
+With `#![feature(trait_alias)]` (RFC #1733), one can define trait aliases, for use in bounds, trait objects, and `impl Trait`. This feature additionaly allows writing `impl` blocks for a subset of trait aliases.
+
+Let's rewrite our AFIT example from before, in terms of this feature. Here's what it looks like now:
+
+```rust
+//! crate frob-lib
+#![feature(trait_alias)]
+
+pub trait LocalFrobber {
+    async fn frob(&self);
+}
+
+pub trait Frobber = LocalFrobber<frob(..): Send> 
+where
+    // not `+ Send`!
+    Self: Send;
+```
+
+```rust
+//! crate joes-crate
+#![feature(trait_alias_impl)]
+
+use frob_lib::Frobber;
+
+struct MyType;
+
+impl Frobber for MyType {
+    async fn frob(&self) {
+        println!("Sloo is 120% klutzed. Initiating brop sequence...")
+    }
+}
+```
+
+Joe's original code Just Works.
+
+The rule of thumb is: if you can copy everything between the `=` and `;` of a trait alias, and paste it between the
+`for` and `{` of a trait `impl` block, and the result is sytactically valid—then the trait alias is implementable.
+
+# Reference-level explanation
+
+A trait alias has the following syntax (using the Rust Reference's notation):
+
+> [Visibility](https://doc.rust-lang.org/stable/reference/visibility-and-privacy.html)<sup>?</sup> `trait` [IDENTIFIER](https://doc.rust-lang.org/stable/reference/identifiers.html) [GenericParams](https://doc.rust-lang.org/stable/reference/items/generics.html)<sup>?</sup> `=` [TypeParamBounds](https://doc.rust-lang.org/stable/reference/trait-bounds.html)<sup>?</sup> [WhereClause](https://doc.rust-lang.org/stable/reference/items/generics.html#where-clauses)<sup>?</sup> `;`
+
+For example, `trait Foo<T> = PartialEq<T> + Send where Self: Sync;` is a valid trait alias.
+
+Implementable trait aliases must follow a more restrictive form:
+
+> [Visibility](https://doc.rust-lang.org/stable/reference/visibility-and-privacy.html)<sup>?</sup> `trait` [IDENTIFIER](https://doc.rust-lang.org/stable/reference/identifiers.html) [GenericParams](https://doc.rust-lang.org/stable/reference/items/generics.html)<sup>?</sup> `=` [TypePath](https://doc.rust-lang.org/stable/reference/paths.html#paths-in-types) [WhereClause](https://doc.rust-lang.org/stable/reference/items/generics.html#where-clauses)<sup>?</sup> `;`
+
+For example, `trait Foo<T> = PartialEq<T> where Self: Sync;` is a valid implementable alias. The `=` must be followed by a single trait (or implementable trait alias), and then some number of where clauses.
+
+An impl block for a trait alias looks just like an impl block for the underlying trait. The alias's where clauses
+are treated as obligations that the the `impl`ing type must meet.
+
+```rust
+pub trait CopyIterator = Iterator<Item: Copy> where Self: Send;
+
+struct Foo;
+
+impl CopyIterator for Foo {
+    type Item = i32; // Would be an error if this was `String`
+
+    fn next(&mut self) -> Self::Item {
+        42
+    }
+}
+
+struct Bar;
+impl !Send for Bar;
+
+// ERROR: `Bar` is not `Send`
+// impl IntIterator for Bar { /* ... */ }
+```
+
+If the trait alias uniquely constrains a portion of the `impl` block, that part can be omitted.
+
+```rust
+pub trait IntIterator = Iterator<Item = i32> where Self: Send;
+
+struct Baz;
+
+impl IntIterator for Baz {
+    // The alias constrains `Self::Item` to `i32`, so we don't need to specify it
+    // (though we are allowed to do so if desired).
+    // type Item = i32;
+
+    fn next(&mut self) -> i32 {
+        -27
+    }
+}
+```
+
+Trait aliases also allow omitting implied `#[refine]`s:
+
+```rust
+//! crate frob-lib
+#![feature(trait_alias)]
+
+pub trait LocalFrobber {
+    async fn frob(&self);
+}
+
+// not `+ Send`!
+pub trait Frobber = LocalFrobber<frob(..): Send> where Self: Send;
+```
+
+```rust
+//! crate joes-crate
+#![feature(trait_alias_impl)]
+
+use frob_lib::Frobber;
+
+struct MyType;
+
+impl Frobber for MyType {
+    // The return future of this method is implicitly `Send`, as implied by the alias.
+    // No `#[refine]` is necessary.
+    async fn frob(&self) {
+        println!("Sloo is 120% klutzed. Initiating brop sequence...")
+    }
+}
+```
+
+Trait aliases are `unsafe` to implement iff the underlying trait is marked `unsafe`.
+
+# Drawbacks
+
+- The sytactic distance between implementable and non-implementable aliases is short, which might confuse users.
+In particular, the fact that `trait Foo = Bar + Send;` means something different than `trait Foo = Bar where Self: Send;` will likely be surprising to many.
+- Adds complexity to the language.
+- Many of the motivating use-cases involve language features that are not yet stable, or even merely speculative.
+More experience with those features might unearth better alternatives.
+
+# Rationale and alternatives
+
+- Very lightweight, with no new syntax forms. Compare "trait transformers" proposals, for example—they are generally much heavier.
+- Better ergonomics compared to purely proc-macro based solutions.
+- One alternative is to allow marker traits or auto traits to appear in `+` bounds of implementable aliases.
+(For example, `trait Foo = Bar + Send;` could be made implementable). However, I suspect that the complexity would not be worthwhile.
+- Another possibility is to require an attribute on implmenentable aliase; e.g. `#[implementable] trait Foo = ...`. Again, I don't think that the complexity is warranted.
+
+# Prior art
+
+- [`trait_transformer` macro](https://github.com/google/impl_trait_utils)
+
+# Unresolved questions
+
+- How does `rustdoc` render these? Consider the `Frobber` example—ideally, `Frobber` should be emphasized
+compared to `LocalFrobber`, but it's not clear how that would work.
+
+# Future possibilities
+
+- New kinds of bounds: anything that makes `where` clauses more powerful would make this feature more powerful as well.
+  - Variance bounds would allow this feature to support backward-compatible GATification.
+  - Method unsafety bounds would support the `Future` → `Async` use-case.