|
| 1 | +- Feature Name: impl-only-use |
| 2 | +- Start Date: 2017-10-01 |
| 3 | +- RFC PR: https://github.com/rust-lang/rfcs/pull/2166 |
| 4 | +- Rust Issue: https://github.com/rust-lang/rust/issues/48216 |
| 5 | + |
| 6 | +# Summary |
| 7 | +[summary]: #summary |
| 8 | + |
| 9 | +The `use …::{… as …}` syntax can now accept `_` as alias to a trait to only import the |
| 10 | +implementations of such a trait. |
| 11 | + |
| 12 | +# Motivation |
| 13 | +[motivation]: #motivation |
| 14 | + |
| 15 | +Sometimes, we might need to `use` a trait to be able to use its methods on a type in our code. |
| 16 | +However, we might also not want to import the trait symbol (because we redefine it, for instance): |
| 17 | + |
| 18 | +```rust |
| 19 | +// in zoo.rs |
| 20 | +pub trait Zoo { |
| 21 | + fn zoo(&self) -> u32; |
| 22 | +} |
| 23 | + |
| 24 | +// several impls here |
| 25 | +// … |
| 26 | +``` |
| 27 | + |
| 28 | +```rust |
| 29 | +// in main.rs |
| 30 | +struct Zoo { |
| 31 | + // … |
| 32 | +} |
| 33 | + |
| 34 | +fn main() { |
| 35 | + let x = "foo"; |
| 36 | + let y = x.zoo(); // won’t compile because `zoo::Zoo` not in scope |
| 37 | +} |
| 38 | +``` |
| 39 | + |
| 40 | +To solve this, we need to import the trait: |
| 41 | + |
| 42 | +```rust |
| 43 | +// in main.rs |
| 44 | +use zoo::Zoo; |
| 45 | + |
| 46 | +struct Zoo { // wait, what happens here? |
| 47 | + // … |
| 48 | +} |
| 49 | + |
| 50 | +fn main() { |
| 51 | + let x = "foo"; |
| 52 | + let y = x.zoo(); |
| 53 | +} |
| 54 | +``` |
| 55 | + |
| 56 | +However, you can see that we’ll hit a problem here, because we define an ambiguous symbol. We have |
| 57 | +two solutions: |
| 58 | + |
| 59 | +- Change the name of the `struct` to something else. |
| 60 | +- Qualify the `use`. |
| 61 | + |
| 62 | +The problem is that if we qualify the `use`, what name do we give the trait? We’re not even |
| 63 | +referring to it directly. |
| 64 | + |
| 65 | +```rust |
| 66 | +use zoo::Zoo as ZooTrait; |
| 67 | +``` |
| 68 | + |
| 69 | +This will work but seems a bit like a hack because rustc forces us to give a name to something we |
| 70 | +won’t use in our types. |
| 71 | + |
| 72 | +This RFC suggests to solve this by adding the possibility to explictly state that we won’t directly |
| 73 | +refer to that trait, but we want the impls: |
| 74 | + |
| 75 | +```rust |
| 76 | +use zoo::Zoo as _; |
| 77 | +``` |
| 78 | + |
| 79 | +# Guide-level explanation |
| 80 | +[guide-level-explanation]: #guide-level-explanation |
| 81 | + |
| 82 | +Qualifying a `use` with `_` on a trait imports the trait’s `impl`s but not the symbol directly. It’s |
| 83 | +handy if you don’t use the trait’s symbol in your type and if you redefine the symbol to something |
| 84 | +else. |
| 85 | + |
| 86 | +The `_` means that you “don’t care about the name rustc will use for that qualified `use`“. |
| 87 | + |
| 88 | +# Reference-level explanation |
| 89 | +[reference-level-explanation]: #reference-level-explanation |
| 90 | + |
| 91 | +`use Trait as _` needs to desugar into `use Trait as SomeGenSym`. With this scheme, global imports |
| 92 | +and exports can work properly with such items, i.e. import / re-export them. |
| 93 | + |
| 94 | +```rust |
| 95 | +mod m { |
| 96 | + pub use Trait as _; |
| 97 | + |
| 98 | + // `Trait` is in scope |
| 99 | +} |
| 100 | + |
| 101 | +use m::*; |
| 102 | + |
| 103 | +// `Trait` is in scope too |
| 104 | +``` |
| 105 | + |
| 106 | +In the case where the symbol is not a *trait*, it works the exact same way. However, a warning must |
| 107 | +be emitted by the compiler to state the unused import (as types don’t have `impl`!). |
| 108 | + |
| 109 | +In the same way, it’s possible to use the same mechanism with `extern crate` for linking-only |
| 110 | +crates: |
| 111 | + |
| 112 | +```rust |
| 113 | +extern crate my_crate as _; |
| 114 | +``` |
| 115 | + |
| 116 | +# Drawbacks |
| 117 | +[drawbacks]: #drawbacks |
| 118 | + |
| 119 | +This RFC tries to solve a very specific problem (when you *must* alias a trait use). It’s just a |
| 120 | +nit to make the syntax more *“rust-ish”* (it’s very easy to think such a thing would work given the |
| 121 | +way `_` works pretty much everywhere else). |
| 122 | + |
| 123 | +# Rationale and alternatives |
| 124 | +[alternatives]: #alternatives |
| 125 | + |
| 126 | +The simple alternative is to let the programmer give a name to the qualified import, which is not a |
| 127 | +big deal, but is a bit ugly. |
| 128 | + |
| 129 | +# Unresolved questions |
| 130 | +[unresolved]: #unresolved-questions |
0 commit comments