Why do Primary Constructors have a limitation that they must be a record class/struct to have a body that is just a semicolon?

[RoyAwesome]

I was reading the Primary Constructor proposal, and this line jumped out at me:

It is an error for a class_body or struct_body to consist of just a ; unless the corresponding class_declaration or struct_declaration has a record keyword and a parameter_list.

Why this restriction? Some

struct Foo(int x, int y);

seems useful.

[DavidArno]

See #6569. Primary constructors are still in design and hopefully will make it into the language soon(ish).

As @Richiban mentions below, under the current proposal for how primary constructors will work, struct Foo(int x, int y); won't produce anything particularly useful.

Due to C# being open source these days though, it's possible to play around with the current prototype, eg via the Primary Constructors branch in SharpLab. This link highlights the difference in terms of code that you'll get with the following too structs:

struct Foo(int x, int y);

struct Bar(int x, int y)
{
    public int X {get;} = x;
    public int Y {get;} = y;
}

[Richiban]

@RoyAwesome

struct Foo(int x, int y);

As far as I'm aware, this would be a useless type because in non-record types the class parameters function as captures (that can be lifted into fields if required by the class members). That means that your type struct Foo(int x, int y); is functionally equivalent to:

struct Foo
{    
    public Foo(int x, int y) 
    {
    }
}

I can see an argument that it should be a warning rather than error, but it's still useless.

[HaloFour]

Correct, primary constructors on non-record types do not synthesize any members. They are only captures which may be referenced by members within the type.

However, they could be used with a source generator which would emit members, which might be helpful for those who want different code generation than is provided by records.

[RoyAwesome]

@Richiban Thanks for the reply. That's a good reason I missed in my reading. I was thinking they synthesized members, but given that it does not, that makes sense!

However, they could be used with a source generator which would emit members, which might be helpful for those who want different code generation than is provided by records.

Perhaps partial classes with just a ; for a body could also not be an error?

[333fred]

Thanks for bringing this up. We'll be discussing the restriction in LDM soon (likely next week, but no guarantees).

[333fred]

Discussed in LDM: https://github.com/dotnet/csharplang/blob/main/meetings/2023/LDM-2023-01-18.md#semi-colon-bodies-for-type-declarations