Why overloading does not occur?

Question

I have the following class:

class CrmToRealTypeConverter : IConverter
{
    #region IConverter Members

    public object Convert<T>(T obj)
    {
        return Convert(obj);
    }

    #endregion

    private DateTime? Convert(CrmDateTime obj)
    {
        return obj.IsNull == false ? (DateTime?)obj.UserTime : null;
    }

    private int? Convert(CrmNumber obj)
    {
        return obj.IsNull == false ? (int?)obj.Value : null;
    }

    private decimal? Convert(CrmDecimal obj)
    {
        return obj.IsNull == false ? (decimal?)obj.Value : null;
    }

    private double? Convert(CrmDouble obj)
    {
        return obj.IsNull == false ? (double?)obj.Value : null;
    }

    private float? Convert(CrmFloat obj)
    {
        return obj.IsNull == false ? (float?)obj.Value : null;
    }

    private decimal? Convert(CrmMoney obj)
    {
        return obj.IsNull == false ? (decimal?)obj.Value : null;
    }

    private bool? Convert(CrmBoolean obj)
    {
        return obj.IsNull == false ? (bool?)obj.Value : null;
    }
}

I am trying to specialize the Convert method with concreate types.
Currently it just loops recursively in Convert<T>() until a stack overflow occurs.

Answer 1

Polymorphism doesn't work on arguments to a method call. An approach you can use it to check the type of obj, cast it to the specific type and then call the appropriate overload.

public object Convert(object obj)
{
    if (obj is CrmDateTime)
        return Convert((CrmDateTime)obj);
    if (obj is CrmNumber)
        return Convert((CrmNumber)obj);
    // ...
}

Answer 2

Late-binding doesn't happen the way you think it does; the compiler binds the call to Convert(obj) in thepublic object Convert<T>(T obj) method to the same method (recursive call). The behaviour you appear to be expecting is that the CLR will dynamically choose the most appropriate overload to execute at run-time, but it doesn't work that way. Try something like this instead:

public object Convert<T>(T obj)
{
   if (obj == null)
       throw new ArgumentNullException("obj");

    var cdt = obj as CrmDateTime;   
    if (cdt != null)
        return Convert(cdt); // bound at compile-time to DateTime? Convert(CrmDateTime)

    var cn = obj as CrmNumber;    
    if (cn != null)
        return Convert(cn); // bound at compile-time to int? Convert(CrmNumber)

    // ...    

    throw new NotSupportedException("Cannot convert " + obj.GetType());
}

If you prefer, you can use reflection here. Such a solution would look something like:

// Making the method generic doesn't really help
public object Convert(object obj) 
{
   if (obj == null)
       throw new ArgumentNullException("obj");

    // Target method is always a private, instance method
    var bFlags = BindingFlags.Instance | BindingFlags.NonPublic;

    // ..which takes a parameter of the obj's type.      
    var parameterTypes = new[] { obj.GetType() };

    // Get a MethodInfo instance that represents the correct overload
    var method = typeof(CrmToRealTypeConverter)
                 .GetMethod("Convert", bFlags, null, parameterTypes, null);

    if (method == null)
        throw new NotSupportedException("Cannot convert " + obj.GetType());

    // Invoke the method with the forwarded argument
    return method.Invoke(this, new object[] { obj });
}  

Answer 3

The model you should follow is the model in the .Net Convert class, there is no reason for you to make the constructor a generic, it brings nothing to the table. Change the convert routines to static methods and the class itself to static:

static class CrmToRealTypeConverter : IConverter
{
    #region IConverter Members

    public static DateTime? Convert(CrmDateTime obj)
    {
        return obj.IsNull == false ? (DateTime?)obj.UserTime : null;
    }

    public static int? Convert(CrmNumber obj)
    {
        return obj.IsNull == false ? (int?)obj.Value : null;
    }

    public static decimal? Convert(CrmDecimal obj)
    {
        return obj.IsNull == false ? (decimal?)obj.Value : null;
    }

    public static double? Convert(CrmDouble obj)
    {
        return obj.IsNull == false ? (double?)obj.Value : null;
    }

    public static float? Convert(CrmFloat obj)
    {
        return obj.IsNull == false ? (float?)obj.Value : null;
    }

    public static decimal? Convert(CrmMoney obj)
    {
        return obj.IsNull == false ? (decimal?)obj.Value : null;
    }

    public static bool? Convert(CrmBoolean obj)
    {
        return obj.IsNull == false ? (bool?)obj.Value : null;
    }
}

Then when you call one of the convert methods the compiler will select the proper overload to call:

CrmDateTime crmDate;
CrmToRealTypeConverter.Convert(crmDate);  // Will call the static DateTime? Convert(CrmDateTime obj) overload    
// or 
CrmNumber crmNum;
CrmToRealTypeConverter.Convert(crmNum);  // Will call the static int? Convert(CrmNumber obj) overload
// and so on...

Edit: If you do the following:

CrmFloat num;
// ...
Object obj = num;
CrmToRealTypeConverter.Convert(obj);

it will not work as the compiler doesn't know the type to match the overload. You would have to cast it and it will work:

CrmToRealTypeConverter.Convert((CrmFloat)obj);

Answer 4

This occurs because the compiler doesn't know the generic type of T until runtime and binds the call to T = System.Object at compile-time, and the only function suitable to take a System.Object is that function itself. However, in .NET 4, you can use the dynamic keyword to cause the runtime to select the correct overload dynamically based on T at runtime, which is what you are looking to happen.

Simple example:

class Main {
    static void somefunction(System.String str)
    {
        System.Console.WriteLine("In String overload");
    }
    static void somefunction(System.Object obj)
    {
        System.Console.WriteLine("In Object overload");
    }
    static void somegenericfunction<T>(T object)
    {
        somefunction(object);
    }
    static void dynamicfunction<T>(dynamic T object)
    {
        somefunction(object);
    }
    static void main(System.String[] args)
    {
        somegenericfunction("A string"); // Calls Object overload, even though it's a String.
        dynamicfunction("A string"); // Calls String overload
    }
}

Note that I don't actually have my compiler on hand and this might not compile literally, but close enough.