Expand the [NotNull...] attributes to cover collections

[SRNissen]

Consider a method static boool AnyNull(...) used like:

    object? NullForgivingMethod(object? a, object? b object? c)
    {
        if(AnyNull(a,b,c) return null; // <-- here
        else              return NullEnforcingMethod(a,b,c);
    }

Here, if AnyNull has overloads for 1 through X parameters, each parameter can be annotated with [NotNullAttribute].

But if AnyNull is written like (params object?[] objects) there is no way to indicate that each object has been checked for null.

I would like to be able to do that.

[TahirAhmadov]

Unfortunately, this is probably almost impossible. How does C# know that you checked all the elements? It now has to interpret a loop's conditions - like for(int i = 0; i < objects.Length; ++i) { ... } - then foreach(var x in objects) { ... } - then how to interpret what's inside the { ... }?

[SRNissen]

I don't know what you're getting at - it seems to work as expected when I test it?

    //In class FF
    public static bool AnyNull([NotNullWhen(false)] object? obj1,
                               [NotNullWhen(false)] object? obj2,
                               [NotNullWhen(false)] object? obj3)
    {
        return false;
    }

    //In test project
    [Test]
    public void AnyNull()
    {
        string? str = null;
        List<object>? list = null;
        object? obj = null;

            if (FF.AnyNull(str, list, obj)) return;
		
        list.Add(str); // throws System.NullReferenceException as expected, ending the test
        list.Add(obj);

        Assert.That(list, Has.Count.EqualTo(2));
    }

Despite compiling at /W4, this code has no warnings when using the AnyNull implementation you provided, exactly like [NotNullWhen(false)] advertises.

[TahirAhmadov]

Strange, it works differently when ref is specified on these parameters. Perhaps somebody who maintains the analyzer around NotNullWhenAttribute can pitch in?

[SRNissen]

"Huh."

Compiling example to work from: https://godbolt.org/z/5r354jbfq

using System;
using System.Diagnostics.CodeAnalysis;

class Program
{
    public static bool AnyNull([NotNullWhen(false)] object? obj1,
                               [NotNullWhen(false)] object? obj2)
    {
        return false;
    }

    static void Main()
    {
        string? greeting = "Hello";
        string? subject = "world";

        if(!AnyNull(greeting, subject))
            Console.WriteLine(greeting.Length + subject.Length);

        greeting = null;
        subject = null;

        if(!AnyNull(greeting, subject))
            Console.WriteLine(greeting.Length + subject.Length);
    }
}

[TahirAhmadov]

Look at this:
https://sharplab.io/#v2:EYLgtghglgdgPgWAFAAEAMACFBGAdAESggHMYB7AZwBcoBjC3AYTIBMBTAQRggBsBPClAoBuZMhQBmLACYMjDAG9kGFVikoALBgCyACgCUi5aoC+xlZKzYAbBmBkyPDFz4A5AK48eugNquyVB5eAOoAFmwwulQATu5s+gC6GNFsAGYYZMAAVmy0VAD8GdnYADTmqhWVVdWqfgFBPGERUbHxSSnpmTl5hV3SZUg1Q8MVdYGejeGRMXGJyWlF3QWLEvrlSoPVUKm6XdgYALwHGDAThigA7BgzbKKbVdu72bJHJ2dYVzd3NY9dUq+nLznT6tb7VS4YVK8Ci3cpmJDIExAA=

[SRNissen]

That's... interesting.

I'm on dotnet 7, both with the code I'm writing that made me start this feature request, and on the Godbolt example, my compiler is (checks...) Roslyn, and I'm on debug mode with warning level 4 and warnings as errors.

Let me try changing some of those around, see if that makes a difference.

EDIT: No I can't seem to get it to warn on this. Keyword ref might make a real difference here?

EDIT 2: Alright - rather than getting non-ref to warn, instead I tried getting ref to not warn and that's doable. You go back a few minor versions and it stops warning. I think the lesson here is: If the analyzer can tell that your annotation is bogus, it will warn you. If your code is too complex for the analyzer to tell, it will trust you to use the annotations correctly.

If so, that would be fine by me - I already only need those annotations for situations where the analyzer cannot tell that the null check has been performed.

EDIT: Dotnet 7, not C# 7.