I don't get it. version(unittest) can't seem to use local variable

[H. S. Teoh via Digitalmars-d-learn]

On Sat, Jul 12, 2014 at 09:20:06PM -0700, Ali Çehreli via Digitalmars-d-learn wrote:
> On 07/12/2014 08:37 PM, H. S. Teoh via Digitalmars-d-learn wrote:
> 
> > ref makes it possible for the caller to modify the pointer returned by
> > the callee. For example:
> >
> > 	class D { int x; }
> > 	class C {
> > 		D d;
> > 		this(D _d) { d = _d; }
> > 		ref D getPtr() { return d; }
> > 	}
> > 	auto d1 = new D;
> > 	auto d2 = new D;
> > 	auto c = new C(d1); // c.d now points to d1
> > 	assert(c.getPtr() is d1); // getPtr returns d1
> > 	c.getPtr() = d2; // this modifies c.d
> > 	assert(c.getPtr() is d2); // getPtr now returns d2
> >
> > If you do not wish the caller to do this, remove the ref from the
> > function signature.
> 
> The twist here is that the OP's function returned 'this' by reference.
> Changing that not only not have any effect on the object, it would
> also be undefined behavior because 'this' is a local variable in that
> use.
> 
> class C {
>     ref C getPtr() { return this; }
> }
> 
> void main()
> {
>     auto c = new C();
>     assert(c.getPtr() is c);
> 
>     c.getPtr() = new C();    // modifying dead variable
>     assert(c.getPtr() is c); // no effect
> }
[...]

Hmm. Shouldn't this be a compiler bug?? Returning a ref to a local
variable should be illegal, even implicit ones like 'this'.


T

-- 
If you want to solve a problem, you need to address its root cause, not just its symptoms. Otherwise it's like treating cancer with Tylenol...