Re: Casts that makes no sense (!)

[Ary Borenszweig]

Steven Schveighoffer wrote:

On Mon, 08 Mar 2010 09:38:08 -0500, Ary Borenszweig <a...@esperanto.org.ar> wrote:

The following compiles:

import std.stdio;

interface I {
}

class A : I {
}

class B {
}

int main() {
   I i = new A();
   A a = cast(A) i;
   B b = cast(B) i; // shouldn't compile
   B c = cast(B) a; // shouldn't compile

   writeln(a);
   writeln(b);
   writeln(c);

   return 0;
}

But two lines there doesn't make sense:

B b = cast(B) i;

An instance of I can never be a B, so why the cast is allowed?

class C: B, I {}

I i = new C;
B b = cast(B)i; // should work

Ah, good one! I just tested it in Java and it also compiles, so I was wrong.

B c = cast(B) a;

An instance of A can never be an A, so why the cast is allowed?

Aside from the typo, I agree with you there, this should never be possible, because a derived class can not inherit both B and A.

Where's the typo?

I think these casts should result in an error. This can prevent some bugs.

Java and C# work like that. You can't cast an object of instance of type A to type B if both types are classes and B isn't a supertype or subtype of A.

This rule makes sense. On the other hand, I wonder how the new ability to cast to any type works in terms of classes. Can you define cast!(A)() in B for instance? I suppose you could make a caveat that either the class must inherit A or define an opCast for A.

Yes, you'd always have to also check if an opCast is present.

-Steve