False negative on nullable analysis

[UniqeId6542]

Version Used:
Visual Studio 18.1.1
C# Tools 5.0.0-2.25574.6

Steps to Reproduce:

internal class DoubleTryGetTest
{
    public static void Test()
    {
        if (TryGet1(out object? o))
        {
            string? s = o.ToString(); // null reference exception produced here.
        }
    }

    private static bool TryGet1([NotNullWhen(true)] out object? o)
    {
        return TryGet2(out o); // no warning
    }

    private static bool TryGet2(out object? o)
    {
        o = null;
        return true;
    }
}

Expected Behavior:
I expect CS8620 to be made somewhere in this code - likely in the body of TryGet1.

Actual Behavior:
No warning is made.

[RikkiGibson]

My recollection is that this is by design. See #48425.

Basically we felt that it was "nuisance-prone" to warn in a scenario like impl of TryGet1, where we don't know the relationship between the return value and the null state of o--versus the case where we do see a relationship (flow state of o appears to depend on the return value), and it's an "incorrect" one.

@jcouv in case you have any further thoughts on this.

[UniqeId6542]

@RikkiGibson, was the issue that false positive warnings were being made?

If TryGet2 was in an assembly that didn't have nullability enabled, I could see not making CS8620. But it was in an assembly with it enabled - so I would expect to be warned that the output may be incorrect.

If I add [NotNullWhen(false)] to TryGet2, I do get a warning. So is it the warning ignored only for ambiguous states? I thought that ambiguous states was a lot of what nullability analysis was for.