Implementing interface method returning byref<'t> cannot just throw

[marklam]

For an interface method with a byref<'t> return, it's not possible to provide an implementation that just throws an exception

Repro steps

open System

type Ii<'T> =
    abstract GetSpan : int -> ReadOnlySpan<'T>

[<Struct>]
type X(value : int) =
    interface Ii<int> with
        member __.GetSpan n = ReadOnlySpan()

[<Struct>]
type Y(value : int) =
    interface Ii<int> with
        member __.GetSpan n = failwith "Nope"

Expected behavior

Compiles

Actual behavior

The line

member __.GetSpan n = failwith "Nope"

reports the error
FS0412 A type instantiation involves a byref type. This is not permitted by the rules of Common IL.

Known workarounds

Include an unreachable dummy return value

        member __.GetSpan n = failwith "Nope";  ReadOnlySpan()

Related information

Visual F# Tools 10.2 for F# 4.5 15.8.0.0. Commit Hash: 39fd7b8.

[cartermp]

Thanks, that definitely looks like a bug.

[dsyme]

It's by design. failwith is a generic function instantiated with the return type. The same would apply to Unchecked.defaultof<_>. Byrefs are just whacky wierd on .NET

You can record a language suggestion if you like

[cartermp]

We'll have to strongly consider doing work here. Since both raise and failwith are generic functions, you cannot throw an exception like you can in C#:

    interface Ii<T>
    {
        int GetSpan(ReadOnlySpan<T> sp);
    }

    struct Y : Ii<int>
    {
        public int GetSpan(ReadOnlySpan<int> sp)
        {
            throw new Exception("Nope");
        }
    }

[dsyme]

Yes, it would make sense to do something here, though it is a bit of a corner case. It's shame .NET generics have this limitation on byref instanatiations.

[marklam]

Could the 'something' also encompass the pipe operator? (like #5286)

[dsyme]

Could the 'something' also encompass the pipe operator? (like #5286)

No. I'm very loathe to go down that long slippery slope since it's a bug farm.

Anything we do here would be for raise and nothing else. Not even for failwith.

[marklam]

No. I'm very loathe to go down that long slippery slope since it's a bug farm.

Fair enough :-)

[cartermp]

@dsyme Do you still think a language suggestion is necessary, or should we just use this to track and update the RFC?

[dsyme]

@dsyme Do you still think a language suggestion is necessary, or should we just use this to track and update the RFC?

We can use this to track I suppose, special casing raise is fine

[cartermp]

Filed: fsharp/fslang-suggestions#872