Any reason why this works?

https://run.dlang.io/is/TALlyw

On Thursday, 4 October 2018 at 07:31:21 UTC, Ritchie wrote:
> Any reason why this works?
>
> https://run.dlang.io/is/TALlyw

Why not?

On Thursday, 4 October 2018 at 07:55:48 UTC, Andrea Fontana wrote:
> On Thursday, 4 October 2018 at 07:31:21 UTC, Ritchie wrote:
>> Any reason why this works?
>>
>> https://run.dlang.io/is/TALlyw
>
> Why not?

The constructor is private.

On Thursday, 4 October 2018 at 07:31:21 UTC, Ritchie wrote:
> Any reason why this works?
>
> https://run.dlang.io/is/TALlyw

"private" applies to the module, not the type. https://dlang.org/spec/attribute.html#visibility_attributes

On Thursday, 4 October 2018 at 07:31:21 UTC, Ritchie wrote:
> Any reason why this works?
>
> https://run.dlang.io/is/TALlyw

Yup.


Alright, so there's a few features in use here - which one are you asking about?

1. Private constructor.
You can call the private constructor because the unit of encapsulation in D is the module, not the type. So everything is visible to everything else inside the same module.

2. @disable this()
This only disables default construction. Since you call a different constructor, this doesn't affect the compilation.

3. @disable this(this)
This disables copy construction. In the example, no copying occurs - the variable declaration leads to a move instead. Simply put, copy construction requires that two copies exist simultaneously, and in this case the potential second copy is immediately destroyed, so no copy is necessary.

4. @disable void opAssign()
There's several reasons this doesn't affect compilation here. First, the signature is wrong - you're disabling an opAssign that takes no arguments. You're probably wanting to do @disable void opAssign(X x);.

Now, even that isn't going to cause it to fail to compile in this case though, since as in #3, what's happening on line 14 is move construction, not assignment. To force an assignment, you need to have an object already:

X x = void;
x = X(5); // calls opAssign

So, now we've explained why it compiles. Currently, there's no way to @disable or hook move construction. There's a DIP in the works (DIP 1014) that aims to provide a hook into move construction, but I don't see that it allows for @disabling it altogether.

Hope this helps!

--
  Simen

On Thursday, 4 October 2018 at 08:14:44 UTC, Simen Kjærås wrote:
> On Thursday, 4 October 2018 at 07:31:21 UTC, Ritchie wrote:
>> Any reason why this works?
>>
>> https://run.dlang.io/is/TALlyw
>
> Yup.
>
>
> Alright, so there's a few features in use here - which one are you asking about?
>
> 1. Private constructor.
> You can call the private constructor because the unit of encapsulation in D is the module, not the type. So everything is visible to everything else inside the same module.
>
> 2. @disable this()
> This only disables default construction. Since you call a different constructor, this doesn't affect the compilation.
>
> 3. @disable this(this)
> This disables copy construction. In the example, no copying occurs - the variable declaration leads to a move instead. Simply put, copy construction requires that two copies exist simultaneously, and in this case the potential second copy is immediately destroyed, so no copy is necessary.
>
> 4. @disable void opAssign()
> There's several reasons this doesn't affect compilation here. First, the signature is wrong - you're disabling an opAssign that takes no arguments. You're probably wanting to do @disable void opAssign(X x);.
>
> Now, even that isn't going to cause it to fail to compile in this case though, since as in #3, what's happening on line 14 is move construction, not assignment. To force an assignment, you need to have an object already:
>
> X x = void;
> x = X(5); // calls opAssign
>
> So, now we've explained why it compiles. Currently, there's no way to @disable or hook move construction. There's a DIP in the works (DIP 1014) that aims to provide a hook into move construction, but I don't see that it allows for @disabling it altogether.
>
> Hope this helps!
>
> --
>   Simen

I apologize for not making it clear. I was talking about the private constructor only. The @disable this() is there to prevent struct literal syntax and the other disables really have no reason to be there for the purpose of this question.

Oddly enough I asked this because I was working on something and was able to call the private constructor from another module, but then I created the run.dlang.io snippet to demo my problem.

I have this in one module: https://pastebin.com/J8r4fyK8

and this in another module:

void testUniquePtr() {
    import bclib.memory.unique_ptr : Unique;

    alias TestUnique = Unique!(int, TestMallocator);
    {
        auto unique = TestUnique(20);
        assert(*unique == 20);
    }
    TestMallocator.check();
}

and it runs without complaints. (SmartPtrMembers doesnt have any constructors in it)

On Thursday, 4 October 2018 at 08:52:50 UTC, Ritchie wrote:
> I apologize for not making it clear. I was talking about the private constructor only. The @disable this() is there to prevent struct literal syntax and the other disables really have no reason to be there for the purpose of this question.
>
> Oddly enough I asked this because I was working on something and was able to call the private constructor from another module, but then I created the run.dlang.io snippet to demo my problem.

Right. Reduced your example to this:

module bar;
struct S() {
    private this()(int n) {}
}

module foo;
unittest {
    import bar;
    auto a = S!()(3);
}

This is apparently a known issue:

https://issues.dlang.org/show_bug.cgi?id=18979

which is basically just a special case of this issue:

https://issues.dlang.org/show_bug.cgi?id=2775

--
  Simen

On Thursday, 4 October 2018 at 09:21:39 UTC, Simen Kjærås wrote:
> This is apparently a known issue:
>
> https://issues.dlang.org/show_bug.cgi?id=18979
>
> which is basically just a special case of this issue:
>
> https://issues.dlang.org/show_bug.cgi?id=2775
>
> --
>   Simen

I see. Looks like the chances of it being fixed anytime soon are slim. Thanks for the info.