When delegate keyword is used in C#, the C# compiler automatically generates a class derived from System.MulticastDelegate class.
This compiler generated class contains 3 methods as well: Invoke, BeginInvoke and EndInvoke.
All these three methods are marked public virtual extern but interestingly the class itself is marked sealed.
Virtual methods defined in a sealed class not only strikes as counter-intuitive but are actually illegal in C#.
So my question is, is there a specific reason for this or is it just one of those harmless things done keeping in mind some hypothetical future enhancement?
Edit 1:
Can the reason be to force use of ‘callVirt’ IL opcode as opposed to ‘call’ so that delegate object is always checked for null by the CLR before trying to execute any of the three methods? Though I fail to see why a delegate should be a special case in this respect.
Also isn’t it a performance hit to force use of callvirt (though it may be minuscule)
Edit 2:
Added CIL tag, as it turns out that the C# way of defining delegates is in fact mandated by the CIL standard. Standard states that (following is not full text)
Delegates shall have a base type of System.Delegate. Delegates shall
be declared sealed, and the only members a delegate shall have are
either the first two or all four methods as specified here. These
methods shall be declared runtime and managed. They shall not have a
body, since that body shall be created automatically by the VES. Other
methods available on delegates are inherited from the class
System.Delegate in the Base Class Library. The delegate methods are:
- The instance constructor
- The Invoke method shall be virtual
- The BeginInvoke method, if present, shall be virtual
- The EndInvoke method shall be virtual
So this is definitely not a side effect of compiler process or is similar to other interesting compiler outputs.
If standard emphasizes something, it must be for some good reason and rationale.
So the question now is why does CIL standard for delegates emphasizes on sealed and virtual at the same time?
Does the catch lie here?:
They shall not have a body, since that body shall be created automatically by the VES.
Are they marked virtual so that the VES/CLR generated body can be executed on invocation of these methods?
As I noted in my question that this sealed virtual anomaly is in fact mandated by CIL standard. It remains unclear why CIL standard specifically mentions that delegate methods
Invoke,BeginInvokeandEndInvokeshould be virtual while at the same time mandating to seal theDelegateinherited class.Also, after going through SSCLI code I learnt that internal optimization of JIT compiler automatically translates any
callvirtcall on a virtual method of a sealed class to normal call with additional null check. This means that delegates do not suffer any performance hit when its Invoke (or any of the other two) method is called throughcallvirtinstruction despite being marked virtual in the IL.When a delegate’s invoke is called, CLR automatically emits a highly optimized body for this method as opposed to compiling IL code to generate body which it does for ‘normal’ methods. This has nothing to do with being marked
virtualin the IL.I have also verified by hand modifying IL code and re-assembling it that virtual can be safely removed from the generated delegate class’s IL code. The generated assembly despite being in violation of the CIL standard runs perfectly fine.
Note that I have converted the virtual methods to normal instance methods.
Since this changed IL runs perfectly fine it proves that standard mandated virtual methods in the sealed delegate class are not necessary. They can be normal instance methods as well.
So in all probability this anomaly is either to emphasize that calling these three delegate methods will in-fact result in calling of some other method (i.e. run-time polymorphism just like ‘normal’ virtual methods) or this has been so to accommodate some future hypothetical enhancement related to delegates.