Assignment and down-casting

[Tim M]

On Sun, 08 Mar 2009 06:59:04 +1300, Yigal Chripun <yigal100 at gmail.com>  
wrote:

> I just read this article by Dennis Richie posted on the NG [1], and I  
> wanted to discuss a way to solve the const return issue.
>
> here's some (general) code:
>
> class A {}
> class B : A {}
>
> A func (A a) {
>    .. do stuff ..
>    return a;
> }
>
> void main() {
>    B b = new B;
>    A a1 = func(b);
>    B a2 = cast(B) func(b); // will not compile with explicit cast
> }
>
> the explicit downcast could be removed if the compiler inserted the  
> down-cast implicitly - assigning null or throwing if the down-cast fails.
>
> this can be applied to constancy since const is super type of both  
> immutable and mutable. I don't know how a const pointer/reference is  
> implemented in D so assume that pointers to const objects have a certain  
> bit set. (constancy runtime type)
>
> // create a const object - bit is set
> const T* cptr = cast(const T*) new Struct;
> // create mutable object - bit unset
> T* ptr = new Struct; // bit is unset
>
> on assingment the bit is preserved and cannot be unset:
> T* a = ptr; // bit unset
> T* b = cptr; // fails - bit cannot be unset
> const T* c = cptr; // bit is set
> const T* d = ptr;  // bit remains unset
>
> the second case fails by either assinging null to be or by throwing an  
> exception.
>
> invariant can use a different bit.
>
> let's say we have:
>
> const T foo(const T a, U b); // foo returns a


Thats bcos it's not a down cast. It's an up cast.
>
> const T cptr = ...; //const object
>        T ptr  = ...; // mutable object
>
> T res = foo (cptr, something); // fails
> T res = foo (ptr, something); // *works*
> const T res = foo (cptr, something); // bit set for res
> const T res = foo (ptr, something); // bit unset for res
>
> just as before, the first foo fails by evaluating to null or by  
> exception and the second foo works since the bit was never set so  
> there's no problem "downcasting" to a regular, mutable, pointer/ref.
>
> What do you think?
>
> [1] http://www.lysator.liu.se/c/dmr-on-noalias.html