LDM Notes for February 26th, 2025

[333fred]

https://github.com/dotnet/csharplang/blob/main/meetings/2025/LDM-2025-02-26.md

Agenda

• Extensions

[timcassell]

4 does mean that we abandon some niceties that we'd get with a new form. For example, it means that we couldn't let type parameters that come from the receiver of an extension method be inferred. However, we have a separate proposal (#8968) for allowing some type parameters to be dropped, and it would be more broadly applicable. Given all of this, we prefer option 4, and will proceed with having new instance extension methods use the same lookup algorithm that current instance extensions use.

That's a shame, I was looking forward to better inference for extension members. How does 8968 address that? I don't see how <, > or <var> helps with extension inference.

[timcassell]

Oops

A contrived example:

public static void DoThing<TElement, TOperator>(this List<TElement> list, TOperator op) where TOperator : IAction<TElement>
    => list.ForEach(e => op.Invoke(e));

You can't call list.DoThing(new MyOperator()) because TElement cannot be inferred.

With new extensions, I would expect it to be fully inferrable (not partial).

extension<TElement>(List<TElement> list)
{
    public void DoThing<TOperator>(TOperator op) where TOperator : IAction<TElement>
        => list.ForEach(e => op.Invoke(e));
}

[Edit] My contrived example is bad because I tried it and it worked. I can't recall the exact kind of extension that I tried where the inference didn't work like I wanted.

[timcassell]

But I have wanted to do something like this.

internal static TTo UnsafeAs<TFrom, TTo>(this TFrom obj) where TFrom : class where TTo : class, TFrom
    => Unsafe.As<TFrom, TTo>(ref obj);

And use it like o.UnsafeAs<MyConcreteType>(). Is that where the partial type inference would come into play? I wouldn't want to have to type o.UnsafeAs<_, MyConcreteType>()

[333fred]

Is that where the partial type inference would come into play?

Correct, yes. You would likely have to write that _, or an empty space with a comma.

[timcassell]

That seems counter unintuitive with the new extension syntax. If I put that method directly on a type, it wouldn't need the extra generic argument.

[333fred]

Yes, that is the comprise.