No nullable warning issued for null event invocation

[stephentoub]

Version Used:
3.1.0-beta1-19172-05+edd2de88fb3e84a097fb30b4070e0f219f624e40

Steps to Reproduce:

#nullable enable
using System;

public class Program
{
    static void Main()
    {
        var p = new Program();
        p.MyEvent(p, EventArgs.Empty);
    }

    public event EventHandler MyEvent;
}

No warning is issued here, but it will null ref.

cc: @dotnet/nullablefc

[HaloFour]

Why would this warn? The event (and the backing field) are both non-nullable. Is the suggestion that the backing field be annotated as nullable if nullability is enabled?

[stephentoub]

There are two possible warnings here. Either:

• it should consider MyEvent to be declaring a non-nullable instance field, in which case this should warn "warning CS8618: Non-nullable field 'MyEvent' is uninitialized.", or
• it should consider MyEvent to be declaring a nullable instance field, in which case this should warn "warning CS8602: Dereference of a possibly null reference."

It's currently doing neither of those.

And it's super rare to initialize an event in a constructor, as 99.9% of the time events are meant to start life without any registrations. So I think declaring an event should map to the latter case, creating the equivalent of a nullable backing field.

But if for some reason we don't want to do that, as noted, this sample should still warn about something.

[HaloFour]

@stephentoub

Good point about the lack of warning on initialization.

Events seem like an odd case to me. As you mentioned the field, by default, is null, and it's unlikely that the class would initialize its own events to anything else. But it makes little sense for the accessors to be nullable.

So I agree, I think even with nullability enabled and the event accessors marked as non-nullable the backing field should still be nullable. I think that would do the most good.

[krwq]

should nullability only affect value in add and remove? Backing field will always be nullable and if declared explicitly then it should be taken care of manually

[jcouv]

We've not implemented nullability analysis of events yet.
Relates to #29839

[stephentoub]

Closing, as the compiler now treats events like fields, and so warns if you've not made the event nullable or initialized it to non-null, and if you make it nullable, then you get invocation warnings.