rust-lang / libs-team

The home of the library team
Apache License 2.0
115 stars 18 forks source link

Soft-destabilize `RustcEncodable` & `RustcDecodable`, remove from prelude in next edition #272

Closed jhpratt closed 11 months ago

jhpratt commented 11 months ago

Proposal

Problem statement

The RustcEncodable and RustcDecoadable derives serve no purpose on their own and are deprecated, undocumented, and widely unused. The derives produce code for rustc-serialize, an external crate which has not had a release since April 2017 and is officially deprecated in favor of serde.

The rustc-serialize crate will, in its current state, fail to compile at some point in the future, as it contains code with a future compatibility warning:

> warning: impl method assumes more implied bounds than the corresponding trait method
>     --> /home/jhpratt/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rustc-serialize-0.3.24/src/serialize.rs:1155:41
>      |
> 1155 |     fn decode<D: Decoder>(d: &mut D) -> Result<Cow<'static, T>, D::Error> {
>      |                                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: replace this type to make the impl signature compatible: `Result<Cow<'a, T>, <D as serialize::Decoder>::Error>`
>      |
>      = warning: this was previously accepted by the compiler but is being phased out; it will become a hard error in a future release!
>      = note: for more information, see issue #105572 <https://github.com/rust-lang/rust/issues/105572>
>      = note: `#[allow(implied_bounds_entailment)]` on by default
> 

Solution sketch

The RustcEncodable and RustcDecodable derives are to be soft-destabilized, as mentioned in rust-lang/rust#92594. This turns any use of them into a deny-by-default lint, which is stronger than the warn-by-default deprecation warning that is currently present.

Additionally, the RustcEncodable and RustcDecodable traits are to be removed from the standard library's preludes in all future editions (i.e. ≥2024). rustfix will change existing uses to use the path to the previous prelude.

Alternatives

A full destabilization is not proposed due to breakage of existing code. If/when rustc-serialize fails to compile on stable as mentioned above, a full destabilization would make more sense.

While this proposes both soft-destabilizing and removing the derives from the prelude, there is no reason one could not be accepted without the other if it is so desired.

Links and related work

dtolnay commented 11 months ago

Seconded; let's make it happen. The soft-destabilize approach described in https://github.com/rust-lang/rust/pull/92594#pullrequestreview-845232777 sounds great to me.