Nullable: CancellationToken

[stephentoub]

No description provided.

[stephentoub]

@jaredpar, I'm confused by the changes that were necessary on this line. Here _kernelEvent is declared as a ManualResetEvent?. As written here:

ManualResetEvent? mre = Interlocked.Exchange<ManualResetEvent?>(ref _kernelEvent!, null)

it compiles without warning.

However, this:

ManualResetEvent? mre = Interlocked.Exchange(ref _kernelEvent, null);

warns on the null argument that Cannot convert null literal to non-nullable reference or unconstrained type parameter and hovering over Exchange to get IntelliSense for it shows that it inferred the type of T to be ManualResetEvent rather than ManualResetEvent?, even though _kernelEvent is ManualResetEvent?. Is it somehow getting messed up by the _kernelEvent != null guard above it?

Similarly, if I change it to be:

ManualResetEvent? mre = Interlocked.Exchange<ManualResetEvent?>(ref _kernelEvent, null);

that warns on the _kernelEvent argument "Possible null reference assignment", which should be fine, given that the type of T here is nullable.

Is this by design or a bug?

[safern]

cc: @gafter @cston

[gafter]

Yes, the null check is causing ref _kernelEvent to be inferred as non-nullable.

[stephentoub]

@jcouv, there are a bunch of places in the framework where we take an Action<object> and an object state to pass to it. Thus, the object in Action<object> callback will be null if and only if state is null. Do we have any attribution planned that will allow for that relationship to be expressed? I don't yet know if it would actually matter much, so just asking right now.

[jcouv]

We have not discussed any such attribute yet.
Feel free to add a comment with scenario details in dotnet/roslyn#26761