I dustmite'd down my problem code, and it seems just changing the order of declarations makes the error go away. Specifically, moving a class declaration from the bottom of the file, to the top, allows it to compile. I also commented on two other things that can be changed to make it work - neither seemingly related to the error at hand.

Just double-checking before I file a bug for this:
--------------------------------------------------
import sumtype;

// DMD 2.089.0-dirty (also v2.087.1)
// app.d(18,13): Error: struct sumtype.SumType!(CallbackType1).SumType is not copyable because it is annotated with @disable

struct Struct1(T) {
    bool[T] items;
}

struct CallbackType1 {
    void delegate(Struct1!Class2) func;   // changing arg to Class2[] compiles OK
}
alias CallbackType = SumType!CallbackType1;

class Class1 {
    CallbackType _callback;
    this(CallbackType callback) { // commenting out this ctor compiles OK
        _callback = callback;
    }
}

// moving Class2 to the top of the file compiles OK
class Class2 {
    Struct1!Class1 subscribers;
}

void main() {}

On 12/8/19 11:43 AM, DanielG wrote:
> I dustmite'd down my problem code, and it seems just changing the order of declarations makes the error go away. Specifically, moving a class declaration from the bottom of the file, to the top, allows it to compile. I also commented on two other things that can be changed to make it work - neither seemingly related to the error at hand.

Yes, if it can compile when you move things around, and the result is *correct* (very important characteristic), then it's definitely a bug. There should be no ordering requirements for module-level code.

-Steve

On Sunday, 8 December 2019 at 18:01:03 UTC, Steven Schveighoffer wrote:
> Yes, if it can compile when you move things around, and the result is *correct* (very important characteristic)

Indeed, everything's working as intended when rearranged.

Thanks!

On Sunday, 8 December 2019 at 18:13:59 UTC, DanielG wrote:
> On Sunday, 8 December 2019 at 18:01:03 UTC, Steven Schveighoffer wrote:
>> Yes, if it can compile when you move things around, and the result is *correct* (very important characteristic)
>
> Indeed, everything's working as intended when rearranged.
>
> Thanks!

Still worth opening an issue.

https://issues.dlang.org/show_bug.cgi?id=20443

On Thursday, 12 December 2019 at 06:23:31 UTC, Basile B. wrote:
> Still worth opening an issue.
>
> https://issues.dlang.org/show_bug.cgi?id=20443

Thanks, Basile. I've added additional information from the github thread where the sumtype author helpfully looked into the problem.