restrict alias this?

[Simen Kjaeraas]

On Sun, 29 Apr 2012 14:28:05 +0200, Namespace <rswhite4 at googlemail.com>  
wrote:

> On Sunday, 29 April 2012 at 11:24:00 UTC, Simen Kjaeraas wrote:
>> On Sun, 29 Apr 2012 10:14:09 +0200, Namespace <rswhite4 at googlemail.com>  
>> wrote:
>>
>>> Based on my previous thread  
>>> (http://forum.dlang.org/thread/rpcqefmyqigftxsgsqui@forum.dlang.org),  
>>> I got the question whether it is possible to restrict "alias this"?
>>> Similar to templates function, e.g. "void foo(T)(T value) if  
>>> (isNumeric!(T) ...", but it should be limited here so that it does not  
>>> interact with the cast or something other.
>>> As far as I know, "alias this" is an implicit cast. And just like an  
>>> explicit cast (opCast) it should be possible to restrict it.
>>>
>>> Otherwise, I have unfortunately run out of ideas how I can allow an  
>>> implicit conversion.
>>
>> What do you mean restrict it?
>
> I mean that you can limit it similar as you can with templates. In my  
> code i'm using "alias this" to enable an implicit conversion from any  
> class, e.g. Foo, to Ref!Foo. Ref is an struct which checks if the value  
> of Foo is null. If it's null a error message will let you know, that you  
> have a problem with one of your objects (i hate the access violation  
> message, that's the original reason for that).
>
> But if i cast with a class, that implements the implicit conversion, as  
> in the following:
>
> [code]
> class A {
> 	mixin TRef!(A);
> }
>
> class B : A { }
>
> A a1 = new B();
>
> B b1 = cast(B) a1.access; // <- problem
> [/code]
>
> the cast will implicit convert with "alias this" to Ref!B. And that  
> isn't what i want and neither what should happen. In this case a limit  
> to "alias this" would be great. A limitation, that "alias this" must not  
> convert the cast to Ref!B.

But that's not what happens. a1's compile-time type is A, so the compiler  
looks
up 'access' in A, but finds nothing. So it tries the alias this, gets a  
Ref!A,
and calls access on that. At this point, T is A, so access returns an A.

Then, the compiler tries to convert that A to a B, and gets confused. This  
is
absolutely a bug. In fact, it is the bug Jesse Phillips reported[1].


[1]: http://d.puremagic.com/issues/show_bug.cgi?id=8001