Why is local struct visible in this outer template, while local class is not?

https://wandbox.org/permlink/MfsDa68qgaMSIr4a

https://dlang.org/spec/template.html#instantiation_scope

---

enum p(T) = __traits(compiles, new T());

class GlobalC {}
struct GlobalS {}

void main()
{
    class LocalC {}
    static assert(p!GlobalC);
    static assert(!p!LocalC);

    struct LocalS {}
    static assert(p!GlobalS);
    static assert(p!LocalS); // ??
}

On 5/11/20 11:11 AM, Shigeki Karita wrote:
> Why is local struct visible in this outer template, while local class is not?
> 
> https://wandbox.org/permlink/MfsDa68qgaMSIr4a
> 
> https://dlang.org/spec/template.html#instantiation_scope
> 
> ---
> 
> enum p(T) = __traits(compiles, new T());
> 
> class GlobalC {}
> struct GlobalS {}
> 
> void main()
> {
>      class LocalC {}
>      static assert(p!GlobalC);
>      static assert(!p!LocalC);
> 
>      struct LocalS {}
>      static assert(p!GlobalS);
>      static assert(p!LocalS); // ??
> }
> 

First, it actually does compile, I think because the compiler recognizes that LocalS is POD (plain old data), without methods, so it doesn't need a context pointer.

e.g.:

auto make(T)() { return new T(); }

...

auto x = make!LocalS; // ok

If you add a method to LocalS, then make!LocalS fails to compile. However, strangely, p!LocalS still returns true, I think that is an error.

-Steve

On Monday, 11 May 2020 at 15:29:53 UTC, Steven Schveighoffer wrote:
> On 5/11/20 11:11 AM, Shigeki Karita wrote:
>> [...]
>
> First, it actually does compile, I think because the compiler recognizes that LocalS is POD (plain old data), without methods, so it doesn't need a context pointer.
>
> e.g.:
>
> auto make(T)() { return new T(); }
>
> ...
>
> auto x = make!LocalS; // ok
>
> If you add a method to LocalS, then make!LocalS fails to compile. However, strangely, p!LocalS still returns true, I think that is an error.
>
> -Steve

Thanks for your answer. Now I understand that POD matters here.

When I add a dtor: `struct LocalS { ~this() {} }`, the `p!LocalS` got false.

On 5/11/20 11:40 AM, Shigeki Karita wrote:
> On Monday, 11 May 2020 at 15:29:53 UTC, Steven Schveighoffer wrote:
>> On 5/11/20 11:11 AM, Shigeki Karita wrote:
>>> [...]
>>
>> First, it actually does compile, I think because the compiler recognizes that LocalS is POD (plain old data), without methods, so it doesn't need a context pointer.
>>
>> e.g.:
>>
>> auto make(T)() { return new T(); }
>>
>> ...
>>
>> auto x = make!LocalS; // ok
>>
>> If you add a method to LocalS, then make!LocalS fails to compile. However, strangely, p!LocalS still returns true, I think that is an error.
>>
> 
> Thanks for your answer. Now I understand that POD matters here.
> 
> When I add a dtor: `struct LocalS { ~this() {} }`, the `p!LocalS` got false.

There is still a bug though. if p!T returns true, but make!T doesn't compile, then something is inconsistent.

-Steve

On Monday, 11 May 2020 at 16:10:36 UTC, Steven Schveighoffer wrote:
> On 5/11/20 11:40 AM, Shigeki Karita wrote:
>> On Monday, 11 May 2020 at 15:29:53 UTC, Steven Schveighoffer wrote:
>>>[...]
>> 
>> Thanks for your answer. Now I understand that POD matters here.
>> 
>> When I add a dtor: `struct LocalS { ~this() {} }`, the `p!LocalS` got false.
>
> There is still a bug though. if p!T returns true, but make!T doesn't compile, then something is inconsistent.
>
> -Steve

I agree. This should be a bug https://wandbox.org/permlink/5YoKKt8xjdexzJHp