Bogus nullable warning for Volatile.Write used in constructor

[jkotas]

See dotnet/runtime#42174 (comment) for full context.

Version Used: 3.8.0-3.20452.3 (.NET SDK 5.0.100-rc.1.20454.5)

Steps to Reproduce:

using System;
using System.Runtime.CompilerServices;
using System.Diagnostics.CodeAnalysis;

class C
{
#nullable enable
    static void X<T>([NotNullIfNotNull("value")] ref T location, T value) where T : class?
    { }

    private readonly double[] foo;

    C()
    {
        if (Environment.CommandLine.Length > 0) foo = new double[5];

        double[] bar = new double[3];
        X(ref foo, bar); // gives CS8601, with `ref foo!`, the warning disappears. 
    }
}

Expected Behavior: No nullable warning

Actual Behavior: Nullable warning

[jkotas]

cc @jcouv

[jcouv]

The proposal under discussion is to have ref foo contribute type double[]? to method type inference (because the value of foo is maybe-null in the invocation) rather than double[] as it does currently.

FYI, I updated the OP example with where T : class?.

Note to self: I'm working on this in branch ref-inference

[am11]

@jcouv, thanks for the correction. Yes it was a typo where T : class? (nullable).

[RikkiGibson]

@jcouv If you're still working on this, could you please self-assign it?

[jcouv]

Fixed in 16.8p4. Note that there was a bit of impact on the runtime repo. It is due to a known issue with inference with null literal (does not properly contribute its nullability) which we're hoping to address in 16.9.