A few years ago I designed a class which had two properties. The two classes ChildA and ChildB had a rich protocol with many properties and methods, but were only distantly related in my model. They both descend from a class DatabaseEntity, which stores a GUID ID. However, there are several different base classes in between in the hierarchy and this relationship is quite trivial. A lot of my classes descend from DatabaseEntity. Over time, two more classes were added to this pattern, so today we have something like this:

public class Parent
{
  public ChildA TheChildA { get; set;}
  public ChildB TheChildB { get; set;}
  public ChildC TheChildB { get; set;}
  public ChildD TheChildB { get; set;}
}

The properties are mutually exclusive: an instance of Parent must have one ONLY of either TheChildA, TheChildB, TheChildC or TheChildD; else these are set to null. The properties are backed by private fields and when one it set to an object, the others are set to null.

My question is this: is it better to plod on as it is, perhaps adding another class to this pattern every 2 years, or refactor and replace the property with a single interface like this:

public class Parent
{
  public IChild TheChild { get; set; }
}

I can then make each of ChildA, ChildB etc implement this interface. It would not be a marker interface; the interface would have perhaps only two members though, and those members wouldn't be that important. In contrast, the Parent and child classes have a hundred or more members.

Does my original pattern represent a particular type of 'code smell'? And does my proposed refactoring conform or represent any particular design pattern or recommended software practice? It strikes me that having maybe two mutually exclusive properties may be acceptable, but having several means the code is quite messy.

A few years ago I designed a class which had two properties. The two classes ChildA and ChildB had a rich protocol with many properties and methods, but were only distantly related in my model. They both descend from a class DatabaseEntity, which stores a GUID ID. However, there are several different base classes in between in the hierarchy and this relationship is quite trivial. A lot of my classes descend from DatabaseEntity. Over time, two more classes were added to this pattern, so today we have something like this:

public class Parent
{
  public ChildA TheChildA { get; set;}
  public ChildB TheChildB { get; set;}
  public ChildC TheChildC { get; set;}
  public ChildD TheChildD { get; set;}
}

The properties are mutually exclusive: an instance of Parent must have one ONLY of either TheChildA, TheChildB, TheChildC or TheChildD; else these are set to null. The properties are backed by private fields and when one it set to an object, the others are set to null.

My question is this: is it better to plod on as it is, perhaps adding another class to this pattern every 2 years, or refactor and replace the property with a single interface like this:

public class Parent
{
  public IChild TheChild { get; set; }
}

I can then make each of ChildA, ChildB etc implement this interface. It would not be a marker interface; the interface would have perhaps only two members though, and those members wouldn't be that important. In contrast, the Parent and child classes have a hundred or more members.

Does my original pattern represent a particular type of 'code smell'? And does my proposed refactoring conform or represent any particular design pattern or recommended software practice? It strikes me that having maybe two mutually exclusive properties may be acceptable, but having several means the code is quite messy.