Uses of trait objects that could be replaced with static polymorphism

[kmcallister]

Not sure how to detect these, but it seems important.

[Manishearth]

So basically suggest the usage of a union type instead of the trait object?

Yeah, I'm not sure what the heuristics should be either.

[jdm]

I think @kmcallister is referring to fn foo<T: SomeTrait>(t: &T) vs. fn foo(t: &SomeTrait) rather than explicit ADTs.

[Manishearth]

Oh, I see.

[coder543]

It seems like the main requirement for this would be to look at every function declaration, such as fn foo(t: &SomeTrait), and then be able to determine that SomeTrait is a trait and not a struct or enum. Once you have that info, you can choose to emit a lint or not.

Maybe the RLS could help with this? not sure.

[oli-obk]

Detecting whether the argument is a trait object is easy. Since any generic type with a bound on a trait can be turned into a trait object of that type, this would be a trivial transformation. The issues with transforming a random function from dynamic dispatch to static dispatch is that someone might be using the function by passing it a trait object. They might not have a chance to do anything else. So we'd also need to add a ?Sized bound.

In gennral you can convert any fn foo(t: &SomeTrait) to fn foo<T: SomeTrait + ?Sized>(t: &T) without breaking anything.

Additionally there are cases where we do not want to use static dispatch in order to reduce compiletimes of downstream crates.