IEnumerable<KeyValuePair<string, string?>> and IEnumerable<KeyValuePair<string, string>>

[TonyValenti]

How do I write a method that can accept an:
IEnumerable<KeyValuePair<string, string?>>
or an
IEnumerable<KeyValuePair<string, string>>
as a parameter value?

Because string is a class, they boil down to the same thing except I can't seem to make the nullable warnings go away.

[huoyaoyuan]

Currently there's no way. It's similar to the problem of Task<T?>.

[jnm2]

I think the problem stems from structs and classes not having generic variance. You can't assign a KeyValuePair<string, string> to KeyValuePair<object, object> or a Task<string> to a Task<object>, and the same invariance holds for KeyValuePair<string, string> -> KeyValuePair<string?, string?>.

[333fred]

That's correct.

[TonyValenti]

It seems odd to be though because a KeyValuePair<string, string> is the same CLR type as a KeyValuePair<string?, string?>. It seems like in a situation like that, a warning shouldn't exist. Especially since a keyvaluepair is essentially all out parameters except for the constructor.

[jnm2]

The trouble is that there's no way to annotate whether TKey and TValue are out or in parameters, and the compiler can't just assume one or the other.

[RikkiGibson]

"Task nullability covariance" #3950 is approved but needs design. I feel that the feature should be extended into a more generalized "nullable variance" feature. KeyValuePair is a particular pain point, and so are tuples and other kinds of generic containers.

[ufcpp]

Real world issue:

using System.Collections.Generic;
using System.Net.Http;

#nullable enable
_ = new FormUrlEncodedContent(new Dictionary<string, string>()); // CS8620 in .NET 5