NRT - Deeper analysis inside constructor for indirect intialization scenarios

[andre-ss6]

The InitializeComponent scenario:

public class Foo {
    public Foo() { // warning here about _x not being initialized
        InitializeComponent();
    }

    public InitializeComponent() {
         _x = "Foo";
         // initialize fields and stuff
    }

    private string _x;
}

Also, happened to me on the first project I'm updating with NRTs:

public class Foo {
    public Foo(Uri baseUrl) {
        baseUrl = null!; // current workaround
        BaseUrl = baseUrl;
    }

    private Uri _baseUrl;
    public Uri BaseUrl { get => _baseUrl; set { /* some logic*/ _baseUrl = value; } }
}

The compiler should follow on the method calls inside a constructor to a reasonable depth in order to determine nullability.

[Joe4evr]

Already pointed this out on gitter, but related to #2328.

[jnm2]

This is coming up frequently enough to be annoying across every codebase (open-source and proprietary) that I've used NRTs with so far. I've been doing = null!.

Flow control could in theory prove that the constructor was calling a method or property accessor which unconditionally initializes that field.