Discussion: Relax type requirements for ??

[333fred]

As part of the implementation for ??= (championed issue here), we're looking at relaxing the requirement on the type of A in ?? (where A is the type of a in a ?? b). Currently, the spec for ?? states:

1. If A exists and is not a nullable type or a reference type, a compile-time error occurs.

We're planning on relaxing this constraint to be:

1. If A exists and is a non-nullable value type, a compile-time error occurs.

This will enable t1 ?? t2, where the type of t1 and t2 are an unconstrained generic parameter, which is currently illegal. We will also enable this for ??=, which will have a similar constraint (for ??=, A must exist).

[orthoxerox]

See also #505, #263

[franklin-ross]

I would also love to see the same relaxation for t1 is null where t1 is an unconstrained generic parameter. It's currently not obvious how to do that effectively and efficiently IMO. The only way I know of is !(value is T), and that looks so strange. Some of the other ways I've seen either likely cause boxing, are not the same, and/or go through the equality functions. Things like (object)default(T) is null, (object)t1 == null, and code fixes like EqualityComparer<T>.Default.Equals(value, default(T)).

Here's a discussion I had over at Roslynator about it. A web search brings up a fair few others like this.

t1 is null very concisely says what it means, avoids boxing, and avoids equality machinery.

[marksmeltzer]

All the things, please 😁

[333fred]

@franklin-ross could you open a separate issue for that? This issue is tracking relaxation for ??. That being said, I think it would be likely @gafter or someone else from the LDM would champion that.

[franklin-ross]

@333fred Yup, sure. Sorry if that was a bit random here, it just seemed related 😅

[gafter]

@franklin-ross We're relaxing things to allow t1 is null in C# 8.0. See #1284. It is already implemented in the preview language version using VS 2019.

[franklin-ross]

I was very pleased to see that when I went to create a new issue, @gafter. Thanks for posting the link here for posterity.