Rfcs: impl PartialEq<Vec<T>> for &'_ [T] / [T; N]

[sollyucko]

Currently:

• Vec<T> implements PartialEq for &'_ [B; N], &'_ [B], &'_ mut [B], and [B; N] (see https://doc.rust-lang.org/std/vec/struct.Vec.html#impl-PartialEq%3C%26%27_%20%5BB%3B%20N%5D%3E)
• [T; N] implements PartialEq for &'b [B], &'b mut [B], [B; N], and [B] (see https://doc.rust-lang.org/std/primitive.array.html#impl-PartialEq%3C%26%27b%20%5BB%5D%3E)
• &'b [B], &'b mut [B], and [B] implement PartialEq<[A; N]> (see https://doc.rust-lang.org/std/primitive.slice.html#impl-PartialEq%3C%5BA%3B%20N%5D%3E)
• [A] implements PartialEq<[B]> (see https://doc.rust-lang.org/std/primitive.slice.html#impl-PartialEq%3C%5BA%3B%20N%5D%3E)

But neither [T; N], &'_ [T], &'_ mut [T], nor [T] implement PartialEq<Vec<T>>. This means that equality tests have to be pointlessly flipped. Is there any reason for this behavior? Would it be possible to fix this?

[shepmaster]

Is there any reason for this behavior?

Yes, the orphan rules. One is defined in libcore, the other is not.

Would it be possible to fix this?

Not really.

---

The RFCs repo isn't really intended for questions. I encourage you to instead search rust-lang/rust repo or the general internet, then post questions on the user's forum or Discord.

[shepmaster]

You can also check out the Rust source code and make the change yourself to see the compiler errors.

[sollyucko]

My change seems to build & pass the tests...:

diff --git a/src/liballoc/vec.rs b/src/liballoc/vec.rs
index b4a9da84787..78dab89be79 100644
--- a/src/liballoc/vec.rs
+++ b/src/liballoc/vec.rs
@@ -2340,6 +2340,8 @@ macro_rules! __impl_slice_eq1 {
 __impl_slice_eq1! { [] Vec<A>, Vec<B>, }
 __impl_slice_eq1! { [] Vec<A>, &[B], }
 __impl_slice_eq1! { [] Vec<A>, &mut [B], }
+__impl_slice_eq1! { [] &[A], Vec<B>, }
+__impl_slice_eq1! { [] &mut [A], Vec<B>, }
 __impl_slice_eq1! { [] Cow<'_, [A]>, &[B], A: Clone }
 __impl_slice_eq1! { [] Cow<'_, [A]>, &mut [B], A: Clone }
 __impl_slice_eq1! { [] Cow<'_, [A]>, Vec<B>, A: Clone }

[shepmaster]

Hmm. You know, there were changes in Rust 1.41 that changed some code in this area. See RFC 2451 for more details. Perhaps it _is_ possible now? If your code compiles, passes tests, and you can write the equality tests you wanted to, it's worth opening a PR to rust-lang/rust.

[sollyucko]

Sounds good!

[mbrubeck]

Would it also be possible to impl PartialEq<[B]> for Vec<A> and PartialEq<Vec<B> for [A]>?

[shepmaster]

@mbrubeck when I tried similar for Option, it caused a massive amount of fallout due to type inference failures. For example, assert_eq!(foo(), None) is no longer unique and requires a turbofish. I assume something similar would happen here.

[mbrubeck]

I can't see any reason that an impl for Vec<A> and [B] would cause inference problems while the existing impl for Vec<A> and &[B] does not.

[sollyucko]

Also, there should be working assert_eq/ne! tests for each possibility, so that shouldn't be an issue.

[mbrubeck]

Would it also be possible to impl PartialEq<[B]> for Vec<A> and PartialEq<Vec<B> for [A]>?

Adding these impls to liballoc/vec.rs appears to work fine, without causing any regressions in the compiler code or tests.

[jplatte]

This can be closed now, right?