Limited type matching?

[Kenji Hara]

On Tuesday, 10 September 2013 at 08:04:48 UTC, Namespace wrote:
>> Currently, all of array types does not allow copy-conversion 
>> like ushort to short for their elements.
>>
>> In old D1 age, static array did not have value semantics, so 
>> the behavior was expected.
>> In D2, static array had changed to value type, but the 
>> behavior was not changed.
>>
>> As far as I know, there was no discussion about that, but at 
>> least it is not fundamentally wrong.
>>
>> Kenji Hara
>
> And what is your personal opinion?
> I think an implicit cast between signed / unsigned would be 
> acceptable even with static arrays. It were only logical.

Hmm, OK. I've taken a look a little deeper about the current 
behavior.

void main()
{
     void f1(short) {}
     void f2(short[2]) {}
     void f3(short[]) {}

     ushort us = 1;
     short ss = us;      // OK
     ss = us;            // OK
     f1(us);             // OK

     ushort[2] usa = [1,2];
     short[2] ssa = usa; // OK -> [x]
     ssa = usa;          // NG -> [y]
     f2(usa);            // NG -> [y]

     ushort[] uda = [1,2];
     short[] sda = uda;  // NG, expected
     sda = uda;          // NG, expected
     f3(uda);            // NG, expected
}

Surely the current state that, [x] is accepted but [y] is 
rejected, looks weird to me. It would be a bug.

Kenji Hara