TypeInfo_Class.interfaces[n].classinfo has TypeInfo_Class and not TypeInfo_Interface?

[Steven Schveighoffer]

On 12/27/17 6:28 PM, Tofu Ninja wrote:
> On Sunday, 24 December 2017 at 05:21:44 UTC, Tofu Ninja wrote:
>> I didn't get any response in learn for this so I will ask it here.
>>
>> TypeInfo_Class.interfaces[n].classinfo has TypeInfo_Class and not 
>> TypeInfo_Interface?
>>
>> Is this correct? Or is it a bug?
>> Doesn't make much sense to me.
>>
>> Also the following code prints false so there are some consequences to 
>> this.
>> # # ]]]]]]]]]]]]
>> import std.stdio;
>> void main(string[] args) {
>>     writeln(typeid(c).interfaces[0].classinfo == typeid(i)); // false
>> }
>> interface i {}
>> class c : i {}
>>
>> What is the proper way to handle this mismatch?
>> Is this mismatch intended or just some implementation detail?
>> Peoples thoughts on this?
> 
> I guess I will just not get an answer to this, seems like just some 
> weirdness of D that will just stick there. The typeinfo system seems 
> really half baked and really provides very little in terms of usefulness.

I'm not even sure why TypeInfo_Interface exists. It seems to be a thin 
wrapper over its TypeInfo_Class member `info`. TypeInfo_Class itself has 
an overridable `info` member which is never overridden, so I'm not sure 
what the purpose of that is either.

Looking at the implementation of TypeInfo_Interface, it appears that the 
only reason to have it, is to allow using interfaces as hash keys. But 
the blunt casting there, I don't think is right. Not all interfaces can 
be cast to Object.

I'll note that opEquals is also implemented incorrectly.

IMO, TypeInfo_Interface should be derived from TypeInfo_Class, and 
simply override the equals/getHash/compare functions. Then change the 
type of 'classinfo' inside the Interface struct to TypeInfo_Interface.

The compiler needs to be updated for this of course.

So in short, I think there are bugs here, but probably not what you 
expected.

-Steve