Cast Action<T> to Action<object> [modified]

[Lost User]

Adriaan Davel wrote:

After much painfull and brain numbing reading I still don't understand why its not possible to cast between Action<T> and Action<object>,

Because T might be an integer, and that's not compatible with object.

I are Troll :suss:

[Adriaan Davel]

Not sure I understand what you mean, this object x = 1; works...

____________________________________________________________ Be brave little warrior, be VERY brave

[Pete OHanlon]

The reason you can't do it is fairly simple. Action<T> is generic - fair enough, and you want it to be Action<object> - ok? Simplistically you would think this would work because no matter what type T was, it would ultimately be convertable to object, as everything derives from object. Suppose though, that type T was a string, and you wanted to use Action<int> - all of a sudden, there's a conversion that is not implicitly possible - in other words, you would have to perform an explicit cast. The cast into object would have to be a special case in the compiler, so they chose not to do this - instead, like for like typing is the only way to achieve this.

I'm not a stalker, I just know things. Oh by the way, you're out of milk.

Forgive your enemies - it messes with their heads

My blog | My articles | MoXAML PowerToys | Onyx

[Adriaan Davel]

Did it build on your side? On my side I get 'Cannot convert type 'System.Func<T>' to 'System.Func<object>'' At least now my build error message is different :)

____________________________________________________________ Be brave little warrior, be VERY brave

[Wayne Gaylard]

Your example would still be an object, allocated on the heap. To use x as an integer would still need to be converted (int)x = ?

[Lost User]

Adriaan Davel wrote:

this object x = 1; works

That wraps your value-type with a reference-type; your number is actually stored in the object, but a number is not an object itself.

I are Troll :suss:

[Adriaan Davel]

Thanks for the reply Pete, just not sure I fully understand your explanation. What you are saying that (assuming this was allowed) this wouldn't be a problem:

Action<object> act = null;
act = new Action<int>();

but the problem would come in if I tried:

act("a");

If that's what you are saying then I'm a bit disappointed in MS because type conversion from object has always posed this challenge (I recon this is why generics was brought in) and yes its fair to expect me to pass the right value in or get a runtime error...

____________________________________________________________ Be brave little warrior, be VERY brave

[Adriaan Davel]

Ok, but in non-generic syntax this works:

private void Button_Click(object sender, RoutedEventArgs e)
{
Calc(1);
}

private void Calc(object val)
{

}

but in generic syntax it doesn't?

____________________________________________________________ Be brave little warrior, be VERY brave

[Lost User]

"Shouldn't"; there's a difference between simple types (anything that's a value) and classes, that's why there's a "where T:class" constraint :)

I are Troll :suss:

[Adriaan Davel]

I know that, and object allows me to pass either and respond accordingly inside the method to which I passed the parameter

____________________________________________________________ Be brave little warrior, be VERY brave

[Lost User]

..because you can check what type of object that it is. That's what VS does at compile-time, and since an integer isn't an object it gives an error. It could be wrapped in an object, that's true - but if that's what should be done, then it has to be explicit. Remember that you can't inherit from an integer, so it does make sense to differentiate between the parameter-behaviour and the generics' behaviour :)

I are Troll :suss:

[Pete OHanlon]

That's exactly what I'm saying. The thing is though, generics were brought in for things like type safety and to prevent aid the developer move away from boxing/unboxing. By having object as the generic type, you've just gone back to having a boxing/unboxing issue - and you're no better off than just passing Action across rather than the generic Action.

I'm not a stalker, I just know things. Oh by the way, you're out of milk.

Forgive your enemies - it messes with their heads

My blog | My articles | MoXAML PowerToys | Onyx

[J4amieC]

Why not just to the following, from what I could see would give youthe exact same result, with the benefit of actually compiling!

public interface IBase
{
Func OnComplete { get; set; }
}

public class Base : IBase
{

public Base(Func onComplete)
{
    OnComplete = onComplete;
}

public Func OnComplete { get; set; }

}
public class GenericBase : IBase
{
public GenericBase(Func onComplete)
{
OnComplete = onComplete;
}

public Func OnComplete { get; set; }

}

[BobJanova]

Generic inheritance is a bit complicated. Here's the reason that you can't cast 'down' the generic type parameter inheritance tree. Imagine this test class:

class BadMistake {
void CompleteHandler(Dictionary<string,int> complexParameter){
// some stuff
}

GenericBase<Dictionary<string,int>> instance;

public void CauseDeath(){
instance = new GenericBase<Dictionary<string,int>>(CompleteHandler);
((Base)instance).OnComplete(16);
}
}

Now, CompleteHandler is expecting to be passed a Dictionary, and you can't compile something which would cause that not to be the case. But using the base class, you can pass an int (or anything else) to it. Method parameter compatibility for generics works backwards. This example will compile if the type parameter in the base is a subclass of T, for example try:

public class Base
{
public Func<T> OnComplete {get;set;} where T:Panel
public Base(Func<T> onComplete) where T:Panel
{
OnComplete = onComplete;
}
}
public class GenericBase<T> : Base
{
public GenericBase(Func<T> onComplete) where T:UserControl : base(onComplete)
{
OnComplete = onComplete;
}
}

(if I got the syntax on the constructor there right). Also, there isn't really a reason to make a non generic base type and inherit from it in this way. You can just use a GenericBase<object> if you want one that can deal with anything.

[Adriaan Davel]

I understand that I should avoid boxing & un-boxing (and I do want to, and I do it everywhere), in this case it's not an option. Bit of a complex scenario I'll have to explain (and the code is part of a 5 000 file solution), but I'm quite irritated that this is disallowed, I thought polymorphism / covariance / contravariance is exactly about this...

____________________________________________________________ Be brave little warrior, be VERY brave

[Lost User]

It is, for objects - aka reference-types, not for value-types. C# is really strict on the difference. The posted alternative with the interface does look like it (could) solve your problem. Alternatively, you could decide only to pass objects, and to wrap your ints in a MyInt class.

I are Troll :suss:

[Rob Philpott]

Because it just can't. If asked questions such as this I like to mutter something about Covariance and Contravariance and wonder off. Usually works in the office environment anyway.

Regards, Rob Philpott.

[RobCroll]

It's a design issue and it makes sense when you think about it. If you want to use a collection of type object, then use ArrayList. The generic List is a strongly typed version of ArrayList, so if strong typing is not required, use ArrayList.

"You get that on the big jobs."

[PIEBALDconsult]

In .net, everything is an object. :-D

[PIEBALDconsult]

Rob Philpott wrote:

Covariance and Contravariance

I thought those were added in .net 4.