调用不同的基本构造函数时如何避免构造函数之间的重复

Question

I have seen many variants of this question asked on SO, but the solutions don't apply in this case.

I have code similar to this, but with lengthy constructors in Derived , which I'd like to avoid duplicating.

abstract class Base
{
    private int ID;

    protected Base( int id )
    {
        ID = id;
    }

    protected Base()
    {
        ID = GenerateID();
    }

    private int GenerateID()
    {
        return 42;
    }
}

class Derived : Base
{
    public readonly int SomeField;
    
    public Derived( int SomeFieldInitialValue ) : base()
    {
        SomeField = SomeFieldInitialValue;
    }

    public Derived( int SomeFieldInitialValue, int id ) : base( id )
    {
        // Want to remove this duplication - could be a long constructor.
        // Can't put it in a method because it does things that only constructors can do.
        SomeField = SomeFieldInitialValue;
    }
}

I can't create a helper method in Derived to contain the code common to all constructors, since some of that code can only be done in constructors (eg setting readonly members).

I can't use just one base constructor and pass along default values, as the value that would get used for Base.ID is known only to Base (computed in the private GenerateID method).

I can't use field initializers, as fields aren't initialised to the same thing in each constructor, and their values depend on constructor parameters.

I'd rather not expose Base.GenerateID to derived classes to pass along in the absence of an id to pass along, as there's no other reason to expose it, and it seems messy to push base functionality up into derived classes. Also there are likely to be other details in future beyond just an int - there's all sorts of other code that may or may not need to be run depending on which Base constructor is called - giving Derived knowledge of that seems bad.

I'd rather avoid removing the readonly from members of Derived as they're readonly on purpose.

I have a feeling that the answer is going to be "sorry, C# doesn't allow that" and I'll have to pick between one of the options I don't want to take :) I think the option that's least bad is to remove readonly or perhaps replace those fields with private set properties.

Answer 1

If you can change the base class, it would be better to have only a single constructor there. This could then provide the two options (fixed ID or newly generated ID) like so:

abstract class Base
{
    private int ID;

    protected Base( int? id )
    {
        ID = id ?? GenerateID();
    }

    private int GenerateID()
    {
        return 42;
    }
}

Also in the derived class, define a (private or protected) constructor doing all the heavy work, and then public constructors redirecting to this.

class Derived : Base
{
    public readonly int SomeField;
    
    protected Derived( int SomeFieldInitialValue, int? id ) : base(id)
    {
        SomeField = SomeFieldInitialValue;
    }
    
    public Derived( int SomeFieldInitialValue ) : this( SomeFieldInitialValue, null)
    {
    }

    public Derived( int SomeFieldInitialValue, int id ) : this( SomeFieldInitialValue, id )
    {
    }
}