I\'m trying add the ability to lookup elements in a List
by overriding the indexer.
using System;
using Sy
Consider this situation:
public class Base
{
public void BaseMethod()
{
}
}
public class Sub : Base
{
public void SubMethod()
{
}
}
public static class Extensions
{
public static void ExtensionMethod(this Base @base) { }
}
Here are some interesting assertions about this code:
ExtensionMethod()
from neither Base
nor Sub
.base.ExtensionMethod()
from Sub
.Extensions.ExtensionMethod(this)
from both Sub
and Base
.this.ExtensionMethod()
from both Sub
and Base
.I don't have a conclusive answer, partly because there might not be one: as you can read in this thread, you have to add this.
if you want to call it in the extension method style.
When you're trying to use an extension method from the type it is in (or - consequently - from a type that is derived from the type used in the extension method), the compiler doesn't realize this and will try to call it as a static method without any arguments.
As the answer states: they [the language designers] felt it was not an important use case scenario to support implicit extension methods (to give the beast a name) from within the type because it would encourage extension methods that really should be instance methods and it was considered plain unnecessary.
Now, it is hard to find out what is happening exactly under the covers but from some playing around we can deduce that base.X()
does not help us. I can only assume that base.X
performs its virtual call as X()
and not this.X()
from the context of the baseclass.
Frankly, I haven't found any truly elegant solution. Consider this scenario:
public class Base
{
protected void BaseMethod()
{
this.ExtensionMethod();
}
}
public class Sub : Base
{
public void SubMethod()
{
// What comes here?
}
}
public static class Extensions
{
public static void ExtensionMethod(this Base @base)
{
Console.WriteLine ("base");
}
public static void ExtensionMethod(this Sub sub)
{
Console.WriteLine ("sub");
}
}
There are 3 ways (leaving aside reflection) to call the ExtensionMethod(Base)
overload:
BaseMethod()
which forms a proxy between the subclass and the extensionmethod. You can use BaseMethod()
, base.BaseMethod()
and this.BaseMethod()
for this since now you're just dealing with a normal instance method which in its turn will invoke the extension method. This is a fairly okay solution since you're not polluting the public API but you also have to provide a separate method to do something that should have been accessible in the context in the first place.
You can also use the primitive way of writing an extension method by skipping the syntactic sugar and going straight to what it will be compiled as. Now you can pass in a parameter so the compiler doesn't get all confused. Obviously we'll pass a casted version of the current instance so we're targetting the correct overload:
Extensions.ExtensionMethod((Base) this);
base.ExtensionMethod()
This is inspired by @Mike z's remark about the language spec which says the following:
At binding-time, base-access expressions of the form
base.I
andbase[E]
are evaluated exactly as if they were written((B)this).I
and((B)this)[E]
, whereB
is the base class of the class or struct in which the construct occurs. Thus,base.I
andbase[E]
correspond tothis.I
andthis[E]
, exceptthis
is viewed as an instance of the base class.
The spec literally says that base.I
will be invoked as ((B) this).I
. However in our situation, base.ExtensionMethod();
will throw a compilation error while ((Base) this).ExtensionMethod();
will work perfectly.
It looks like something is wrong either in the documentation or in the compiler but that conclusion should be drawn by someone with deeper knowledge in the matter (paging Dr. Lippert).
Yes, I would say it is. It kind of feels like a black hole within the C# spec: practically everything works flawlessly but then suddenly you have to jump through some hoops because the compiler doesn't know to inject the current instance in the method call in this scenario.
In fact, intellisense is confused about this situation as well:
We have already determined that that call can never work, yet intellisense believes it might. Also notice how it adds "using PortableClassLibrary" behind the name, indicating that a using
directive will be added. This is impossible because the current namespace is in fact PortableClassLibrary
. But of course when you actually add that method call:
and everything doesn't work as expected.
The main conclusion is simple: it would have been nice if this niche usage of extension methods would be supported. The main argument for not implementing it was because it would encourage people to write extension methods instead of instance methods.
The obvious problem here is of course that you might not always have access to the base class which makes extension methods a must but by the current implementation it is not possible.
Or, as we've seen, not possibly with the cute syntax.