Custom roslyn analyzer to override nullability analysis?

[adamfk]

I understand that a bunch of annotations to inform nullability analysis is already in the works #26761, but I'd like to suggest that it would be useful to allow writing custom roslyn analyzers that can override the compiler's nullability warnings.

Mads mentioned in this video that a mini language of attributes is being developed to inform nullability analysis to cover 98% of cases. I think that custom analyzers might be able to cover the remaining 2% of cases.

The use cases that I'm currently thinking of are initializing fields of an object via reflection during construction.

[jaredpar]

Moving to @dotnet/roslyn-ide as this appears to be more of a feature request for analyzers.

[sharwell]

This would be covered under #20242 if that API moves forward.

[sharwell]

Duplicate of #20242

[Dreamescaper]

@sharwell
Can we reopen this one? I don't think that's same as #20242. Problem is that with such approach I need to re-implement nullability logic that is already present in Roslyn.

Let me give an example from NUnit test.

SomeType? obj = GetSomeObjectWithNullableProperties();
Assert.That(obj?.SomeProperty?.NestedProperty, Is.Not.Null);  <---this line

Suppose I'd like to support nullability changes for Is.Not.Null asserts by analyzer. From that line I need know that:
obj is not null
obj.SomeProperty is not null
obj.SomeProperty.NestedProperty is not null

And to implement that in my suppression analyzer.
But I don't want to do that - Roslyn already knows how to handle that, ideally I'd like to tell Roslyn - consider this expression as not null, and that's it.

[sharwell]

@jaredpar #32042 (comment) sounds like a compiler request

[mavasani]

ideally I'd like to tell Roslyn - consider this expression as not null, and that's it.

Currently no part of the compiler's analysis depends on any third party analyzer's input/output. Introducing any such feedback loop can potentially lead to unwanted deadlocks and/or infinite recursion.