Wrong overload resolution ?

[Vladimir Granitsky]

Hi guys,

Please, look at the code below and try to step into it. The compiled code calls the loosely typed method public void Method1(object o) !?!?

Am I right that C# compiler does wrong overload resolution ?

I've used parameters of type object and string here, just to illustrate the problem. Really I have a bit more deep inheritance graph, and the things get more interesting if the strongly typed overload is like override public void Method1(BaseType x). When I call it with parameter of type SubType (that inherits BaseType) the right method is called.

Thanks for any useful points.

Regadrs,
Vladimir Granitsky
using System;using System.Diagnostics;namespace OverloadResolution{ public class Class1 { virtual public void Method1(string s) { Trace.WriteLine("Class1.Method1"); } } public class Class2 : Class1 { override public void Method1(string s) { Trace.WriteLine("Class2.Method1a"); } public void Method1(object o) { Trace.WriteLine("Class2.Method1b"); } } class Client { [STAThread] static void Main(string[] args) { string s = "blah"; Class2 o2 = new Class2(); o2.Method1(s); } }}

 

[Guest]

Hi Vladimir,

I work with .NET Framework 1.0 has the same "wrong" result!

Only change in "Main":
static void Main(string[] args)
{
string s = "blah";
Class1 o2 = new Class2();
o2.Method1(s);
}

or adding method into Class1, and "override" it in Class2

public virtual void Method1(object o) {
Console.WriteLine("Class1.Method1(object)");
}

gives me "right" result...

It looks like an overriden method has less "priority" than
a non-overriden methods in case of overload.
When i replaced method of Class2:
new public void Method1(string s)
{
Trace.WriteLine("Class2.Method1(string)");
}
then it was used by overload...

Strange and creepy :-(

Marcin

 

[Michael Moreno]

Am I right that C# compiler does wrong overload resolution ?

No.
If you introduce a new definition of an overloaded method then there is
a specific thing to do I cannot remember of right now but it is
documented.

 

[James Curran]

Well, I'm gonna guess that it's working according to the C# spec, but
I'm leaning toward it being a rather bad design choice on the spec writers.

To follow what going on, add the lines:
Class1 o1 = o2;
o1.Method1(s);

to the end of your Main() function. Run it, and you'll note that while
o2.Method1() give the wrong response, o1.Method1() is correct.

Now, replace the "override" with "new" (or just delete it). Now,
o2.Method1() is correct, while o1.Method1() is wrong.

So, what I THOUGHT was happening was that Method1(object) was hiding the
Class1.Method1() (including hiding it's override).
BUT, now change Method1(object o) to Method1(int o). Now, both o2.Method1()
& o1.Method1() are correct. So, it's only hiding it if the parameters are
similar (C++ would hide it based on just the name)


Hi guys,

Please, look at the code below and try to step into it. The compiled code
calls the loosely typed method public void Method1(object o) !?!?

Am I right that C# compiler does wrong overload resolution ?

I've used parameters of type object and string here, just to illustrate the
problem. Really I have a bit more deep inheritance graph, and the things get
more interesting if the strongly typed overload is like override public void
Method1(BaseType x). When I call it with parameter of type SubType (that
inherits BaseType) the right method is called.

Thanks for any useful points.

Regadrs,
Vladimir Granitsky
using System;using System.Diagnostics;namespace OverloadResolution{
public class Class1 { virtual public void Method1(string s)
{ Trace.WriteLine("Class1.Method1"); } } public
class Class2 : Class1 { override public void Method1(string s)
{ Trace.WriteLine("Class2.Method1a"); } public void
Method1(object o) {
Trace.WriteLine("Class2.Method1b"); } } class Client
{ [STAThread] static void Main(string[] args)
{ string s = "blah"; Class2 o2 = new Class2();
o2.Method1(s); } }}

 

[Guest]

Am I right that C# compiler does wrong overload resolution ?

The compiler matches the C# specification here, even though it is surprising
behavior.

The full story of overload resolution is incredibly complicated and it is
hard to reverse engineer the rules by looking at examples. There are two
rules that apply here.

1. Overrides are not included in the set of candidate methods during
resolution. This may seem counterintuitive since an override is more specific
and therefore a better match. However, an override is logically the same
method as the original declared in the base class. The original is still a
candidate after we remove the overrides. If the remaining steps in the
resolution process select the original method (i.e. the virtual version),
dynamic binding will ensure that the correct override version executes.

2. All candidate methods must come from the same type. We use the lowest
type in the hierarchy that contains an applicable method. We would prefer to
select a method from the lowest type since it is the most specific and we
assume it would perform the most appropriate operation. Only if the lowest
type does not contain an applicable method do we look further up the
hierarchy. We keep all the methods declared in that type and discard all
methods declared in all other types. We discard methods in the other types
even if their parameter list is a better match.

 

[Vladimir Granitsky]

Hi James,
Thanks for the interesting response. I agree that the question is - Is this a C# compiler bug or a specification design issue. Will look forward for someone to answer. My comment are below.

Well, I'm gonna guess that it's working according to the C# spec, but
I'm leaning toward it being a rather bad design choice on the spec writers.

To follow what going on, add the lines:
Class1 o1 = o2;
o1.Method1(s);

to the end of your Main() function. Run it, and you'll note that while
o2.Method1() give the wrong response, o1.Method1() is correct.

Yes, I know, This is bacause if the method is declared vurtual in the base class, the latest override will be invoked, even if the object is cast to the base type. Currently I resolve this issue by calling ((Class1)o2).Method1(s);

Now, replace the "override" with "new" (or just delete it). Now,
o2.Method1() is correct, while o1.Method1() is wrong.

I think we can't say wrong here. The "new" keyword prevents Class2.Method1(string s) from being an override of Class1.Method1(string s) and the rule i mentioned above do not apply. So if you have a variable of type Class1 pointing to an instance of Class2, the method of Class1 will be invoked. I think, this is normal behaviour.

So, what I THOUGHT was happening was that Method1(object) was hiding the
Class1.Method1() (including hiding it's override).

BUT, now change Method1(object o) to Method1(int o). Now, both o2.Method1()
& o1.Method1() are correct. So, it's only hiding it if the parameters are
similar (C++ would hide it based on just the name)

I think this is because string cannot cast to int and the compiler takes the right way.

Hi guys,

Please, look at the code below and try to step into it. The compiled code
calls the loosely typed method public void Method1(object o) !?!?

Am I right that C# compiler does wrong overload resolution ?

I've used parameters of type object and string here, just to illustrate the
problem. Really I have a bit more deep inheritance graph, and the things get
more interesting if the strongly typed overload is like override public void
Method1(BaseType x). When I call it with parameter of type SubType (that
inherits BaseType) the right method is called.

Thanks for any useful points.

Regadrs,
Vladimir Granitsky
using System;using System.Diagnostics;namespace OverloadResolution{
public class Class1 { virtual public void Method1(string s)
{ Trace.WriteLine("Class1.Method1"); } } public
class Class2 : Class1 { override public void Method1(string s)
{ Trace.WriteLine("Class2.Method1a"); } public void
Method1(object o) {
Trace.WriteLine("Class2.Method1b"); } } class Client
{ [STAThread] static void Main(string[] args)
{ string s = "blah"; Class2 o2 = new Class2();
o2.Method1(s); } }}

 

[Bruce Wood]

2. All candidate methods must come from the same type. We use the
lowest

type in the hierarchy that contains an applicable method. We would prefer to
select a method from the lowest type since it is the most specific and we
assume it would perform the most appropriate operation. Only if the lowest
type does not contain an applicable method do we look further up the
hierarchy. We keep all the methods declared in that type and discard all
methods declared in all other types. We discard methods in the other types
even if their parameter list is a better match.

I find rule #2 jarring... it runs contrary to my understanding of how
inheritance should work, let alone overloading. I had always thought
that if I were to declare:

public class Class1
{
public void Method1(string s) {...}
}

public class Class2 : Class1
{
public void Method1(object o) {..}
}

and called

Class2 c2 = new Class2();
c2.Method1("Hello world");

then of course the method invoked would be the method from the base
class, Class1. You're telling me that the Class2 method would be
invoked instead? That's loopy!

 

[Vladimir Granitsky]

Hi everybody,
At first thank you all for the interesting and reasonable responses.

It seems that compiler beheaves regarding the specification. Yet yesterday I tried to read and understand the spec, but left it after getting headache. As MarkT says "The full story of overload resolution is incredibly complicated ..."

Anyway, do you think that this behaviour is right even if it is correct regarding the specification. Shoud we report this and ask for specification changes ?

In between I tryed to test the case in VB.NET, and just want to let you know that the "problem" does not apear there (You may try the code below).
A colleague of mine says that this "wrong" behaviour is missing in C++.

Cheers,
Vladimir


Class Class1
Public Overridable Sub Method1(ByVal s As String)
Trace.WriteLine("Class1.Method1")
End Sub
End Class

Class Class2
Inherits Class1

Public Overloads Overrides Sub Method1(ByVal s As String)
Trace.WriteLine("Class2.Method1a")
End Sub

Public Overloads Sub Method1(ByVal o As Object)
Trace.WriteLine("Class2.Method1b")
End Sub

End Class

Module Module1
Sub Main()
Dim o2 As Class2 = New Class2
Dim s As String = "blah"
o2.Method1(s)
End Sub
End Module

Hi guys,

Please, look at the code below and try to step into it. The compiled code calls the loosely typed method public void Method1(object o) !?!?

Am I right that C# compiler does wrong overload resolution ?

I've used parameters of type object and string here, just to illustrate the problem. Really I have a bit more deep inheritance graph, and the things get more interesting if the strongly typed overload is like override public void Method1(BaseType x). When I call it with parameter of type SubType (that inherits BaseType) the right method is called.

Thanks for any useful points.

Regadrs,
Vladimir Granitsky
using System;using System.Diagnostics;namespace OverloadResolution{ public class Class1 { virtual public void Method1(string s) { Trace.WriteLine("Class1.Method1"); } }
public class Class2 : Class1 { override public void Method1(string s) { Trace.WriteLine("Class2.Method1a"); }
public void Method1(object o) { Trace.WriteLine("Class2.Method1b"); } }
class Client { [STAThread] static void Main(string[] args) { string s = "blah"; Class2 o2 = new Class2(); o2.Method1(s); } }}

 

[Vladimir Granitsky]

Hi Marcin ,
Thanks for the reply.

Changes in Main() cast the o2 to Class1 and it is normal to get right
result. In fact currently I resolve the problem rigth this way.

adding method into Class1, and "override" it in Class2 also is interesting.
But what if I do not need Method1(object o) in Class 1 ?

As I getting more responses I think that the behaviour is regarding the
spec, but the spec is wrongly witten here.

Regards,
Vladimir

Hi Vladimir,

I work with .NET Framework 1.0 has the same "wrong" result!

Only change in "Main":
static void Main(string[] args)
{
string s = "blah";
Class1 o2 = new Class2();
o2.Method1(s);
}

or adding method into Class1, and "override" it in Class2

public virtual void Method1(object o) {
Console.WriteLine("Class1.Method1(object)");
}

gives me "right" result...

It looks like an overriden method has less "priority" than
a non-overriden methods in case of overload.
When i replaced method of Class2:
new public void Method1(string s)
{
Trace.WriteLine("Class2.Method1(string)");
}
then it was used by overload...

Strange and creepy :-(

Marcin

Hi guys,

Please, look at the code below and try to step into it. The compiled
code calls the loosely typed method public void Method1(object o) !?!?

Am I right that C# compiler does wrong overload resolution ?

I've used parameters of type object and string here, just to illustrate
the problem. Really I have a bit more deep inheritance graph, and the
things get more interesting if the strongly typed overload is
like override public void Method1(BaseType x). When I call it with
parameter of type SubType (that inherits BaseType) the right method is
called.

Thanks for any useful points.

Regadrs,
Vladimir Granitsky

using System;
using System.Diagnostics;
namespace OverloadResolution
{
public class Class1
{
virtual public void Method1(string s)
{
Trace.WriteLine("Class1.Method1");
}
}

public class Class2 : Class1
{
override public void Method1(string s)
{
Trace.WriteLine("Class2.Method1a");
}

public void Method1(object o)
{
Trace.WriteLine("Class2.Method1b");
}
}

class Client
{
[STAThread]
static void Main(string[] args)
{
string s = "blah";
Class2 o2 = new Class2();
o2.Method1(s);
}
}
}

 

[James Curran]

No, that makes sense (It's also how it done in C++). Consider if you
had written this without Class1.Method1. Naturally, you'd expect
Class2.Method1 to be called, as it's the only Method1. Now consider if,
years later, some maintence programmer added Method1 to Class1. Next time
you recompile your code, suddenly a different function is being called.
That's not the way it should work. So, method1 in a derived class hide base
methods with the same name.

 

[Bruce Wood]

That makes more sense, now. I'm still not 100% sold, but I know that
the .NET Framework designers spent a lot of time thinking about
versioning, the contracts implied by class interfaces, and not breaking
derived classes / clients just because a base class changed. That's one
of the reasons why checked exceptions weren't included in C#. So, I can
see them doing this as well in order to solve similar versioning
problems.

It still screws up my notions of class inheritance, though.

 

[Vladimir Granitsky]

I agree. But the case Bruce introduces is different.

 

[Vladimir Granitsky]

Hi,
I agree that candidates must come from the same class. Otherwise we'll fall into the problem described by James.

But I cannot agree that the overrides should be excluded from the candidate list. Also the argument that they are logically same as the original in the base class I think is weak. When a derived class overrides the method in a base class, it states that he know how to perform given operation (even better, with or without the help of the base class). That is the override method should be considered as it comes from the derived class. We cannot find any solid argument that they are not from it.

"Employee methods are preferred over Person methods".

MarkT, I understant what do you mean here, but I think, that through overeload resolution process Employee methods shoud include the overrides too. Otherwise, as it is in fact, the overload resolution cuts off the chance dynamic binding to do it's job.

I have coped with the task to prepare a managed C++ example (included below). The "problem" is missing there. It's looks like only C# is so "smart" in overload resolution. I really like C# and I've never preffered any other language. But this is very concerning.


#include "stdafx.h"
#using <mscorlib.dll>
using namespace System;
using namespace System:iagnostics;

__gc public class CClass1
{
public:
virtual void Method1(String *s)
{
System:iagnostics::Trace::WriteLine("Class1.Method1");
}
};

__gc public class CClass2 : public CClass1
{
public:
void Method1(String *s)
{
System:iagnostics::Trace::WriteLine("Class2.Method1a");
}

void Method1(Object *o)
{
System:iagnostics::Trace::WriteLine("Class2.Method1b");
}
};

int _tmain()
{
CClass2* o2 = new CClass2();
o2->Method1("Hello World!");
return 0;
}

 

[Guest]

I've observed the behaviour you second point refers to, but can't find it in
ECMA334. Do you by any chance know which section it is in?

Thanks,
Aaron.