C# 11: Check that we get AST for structs that doesn't initialise all fields.

[michaelnebel]

This is a test only PR.
In C# 11 it is allowed to declare structs that doesn't initialise all fields/properties.

[tamasvajk]

In IL, the non-initialized fields/auto-properties are explicitly assigned their default value. Do we need to synthesize these implicit assignments in QL?

[michaelnebel]

In IL, the non-initialized fields/auto-properties are explicitly assigned their default value. Do we need to synthesize these implicit assignments in QL?

Good question.
This will require doing some definite assignment analysis, which has to be equivalent to that of the C# compiler itself.
Do you know of any other examples where we do this?

[tamasvajk]

Good question. This will require doing some definite assignment analysis, which has to be equivalent to that of the C# compiler itself. Do you know of any other examples where we do this?

I don't know. I found two PRs that are somewhat related:

• C#: Extract assignments for field/property initializers #1741
• C#: Fix dataflow for default constructors #9769

[michaelnebel]

cc @hvitved : What do you think?

[hvitved]

If we were to synthesize default assignments, that would have to be done in the extractor.

However, we also don't generate default assignments for class fields, so I don't think we want to do it for structs.

[michaelnebel]

If we were to synthesize default assignments, that would have to be done in the extractor.

However, we also don't generate default assignments for class fields, so I don't think we want to do it for structs.

The difference is that with C# 11 it is possible to not explicitly initialize all fields of a struct. If we decides to not assign a value then the compiler will explicitly insert an instruction to do so.
That is for classes,

public class C {
    private readonly int A;
    private readonly int B;
    
    public C() { 
        A = 0;
    }
}

Compiling and de-compiling is the identify, but for following declaration

public struct S {
    private readonly int A;
    private readonly int B;
    
    public S() { 
        A = 0;
    }
}

compiling and de-compiling yields

public struct S
{
    private readonly int A;

    private readonly int B;

    public S()
    {
        B = 0;
        A = 0;
    }
}

[hvitved]

The difference is that with C# 11 it is possible to not explicitly initialize all fields of a struct. If we decides to not assign a value then the compiler will explicitly insert an instruction to do so.

Still, fields of classes also get a default value (i.e., the field B of the class C in your example will have value 0), but we don't synthesize the implicit assignment

private readonly int B = default(int);

[michaelnebel]

The difference is that with C# 11 it is possible to not explicitly initialize all fields of a struct. If we decides to not assign a value then the compiler will explicitly insert an instruction to do so.

Still, fields of classes also get a default value (i.e., the field B of the class C in your example will have value 0), but we don't synthesize the implicit assignment

private readonly int B = default(int);

I am convinced :-)
If you approve the PR, I will merge.