A #[derive] macro for PartialEq implementations that ignore lifetime variance

Consider a type such as

pub struct StrWrap<'a> {
    wrapped: &'a str
}

The standard #[derive(PartialEq)] generates an implementation as follows

impl<'a> PartialEq<StrWrap<'a>> for StrWrap<'a> { ... }

This seems to imply that we can only compare StrWraps where the lifetime of the wrapped strings are identical. Nevertheless, code such as

static STATIC_STR: &'static str = "Hello World!";

let local_str = String::from("Hellow World!");
let wrap1: StrWrap<'static> = StrWrap { wrapped: STATIC_STR };
let wrap2 = StrWrap { wrapped: local_str.as_ref() }; // <-- lets call the anonymous lifetime here '1
assert_eq!(wrap1, wrap2);

compiles and runs just fine, even though the wrapped str's clearly have different lifetimes! Why? The lifetime 'a is covariant: If I have some other lifetime 'b: 'a, then any place that wants a 'a happily accepts a 'b. Or, in our specific case, we see that PartialEq wants two StrWrap<'1>, but since 'static outlives '1, the compiler realises that downgrading the 'static to a '1 for the duration of the call to .eq is perfectly fine. This is called subtyping.

However, what if my lifetime is not covariant? This happens for example with lifetimes used in generic associated types (GATs). In this case, the compiler cannot adjust our lifetimes for us, and the above example will indeed fail to compile, as there is no PartialEq implementation for StrWrap<'1> == StrWrap<'static>, only for StrWrap<'1> == StrWrap<'1> and StrWrap<'static> == StrWrap<'static>. However, being unable to compare two values because the lifetimes are different is a bit silly, especially since lifetimes are not actually a thing at runtime.

We can work around this limitation by hand-rolling our own PartialEq instead of using the build-in derive:

impl<'a, 'b> PartialEq<StrWrap<'b>> for StrWrap<'a> {
    fn eq(&self, other: &StrWrap<'b>) -> bool {
        self.wrapped == other.wrapped
    }
}

With this, our example works again, as now StrWrap<'1> == StrWrap<'static> is possible. Generating these generalized PartialEq implementations is exactly the purpose of this crate.

stadust/variant-partial-eq

A #[derive] macro for PartialEq implementations that ignore lifetime variance