Like Nader said, maybe think about making a more loose-coupled design. However, in my case, loose-coupling was not appropriate. Here's my class hierarchy, and how I propose to solve the problem without using custom serialization or DTOs.
In my project, I'm constructing a whole bunch of objects to represent pieces of an XML document that will be submitted via a web service. There are a very large number of pieces. Not all are sent with every request (actually, in this example, I'm modeling a response, but the concepts are the same). These pieces are used much like building blocks to assemble a request (or disassemble a response, in this case). So here's an example of using aggregation/encapsulation to accomplish the desired ordering despite the inheritance hierarchy.
[Serializable]
public abstract class ElementBase
{
// This constructor sets up the default namespace for all of my objects. Every
// Xml Element class will inherit from this class.
internal ElementBase()
{
this._namespaces = new XmlSerializerNamespaces(new XmlQualifiedName[] {
new XmlQualifiedName(string.Empty, "urn:my-default-namespace:XSD:1")
});
}
[XmlNamespacesDeclaration]
public XmlSerializerNamespaces Namespaces { get { return this._namespaces; } }
private XmlSerializationNamespaces _namespaces;
}
[Serializable]
public abstract class ServiceBase : ElementBase
{
private ServiceBase() { }
public ServiceBase(Guid requestId, Guid? asyncRequestId = null, Identifier name = null)
{
this._requestId = requestId;
this._asyncRequestId = asyncRequestId;
this._name = name;
}
public Guid RequestId
{
get { return this._requestId; }
set { this._requestId = value; }
}
private Guid _requestId;
public Guid? AsyncRequestId
{
get { return this._asyncRequestId; }
set { this._asyncRequestId = value; }
}
private Guid? _asyncRequestId;
public bool AsyncRequestIdSpecified
{
get { return this._asyncRequestId == null && this._asyncRequestId.HasValue; }
set { /* XmlSerializer requires both a getter and a setter.*/ ; }
}
public Identifier Name
{
get { return this._name; }
set { this._name; }
}
private Identifier _name;
}
[Serializable]
public abstract class ServiceResponseBase : ServiceBase
{
private ServiceBase _serviceBase;
private ServiceResponseBase() { }
public ServiceResponseBase(Guid requestId, Guid? asyncRequestId = null, Identifier name = null, Status status = null)
{
this._serviceBase = new ServiceBase(requestId, asyncRequestId, name);
this._status = status;
}
public Guid RequestId
{
get { return this._serviceBase.RequestId; }
set { this._serviceBase.RequestId = value; }
}
public Guid? AsyncRequestId
{
get { return this._serviceBase.AsyncRequestId; }
set { this._serviceBase.AsyncRequestId = value; }
}
public bool AsynceRequestIdSpecified
{
get { return this._serviceBase.AsyncRequestIdSpecified; }
set { ; }
}
public Identifier Name
{
get { return this._serviceBase.Name; }
set { this._serviceBase.Name = value; }
}
public Status Status
{
get { return this._status; }
set { this._status = value; }
}
}
[Serializable]
[XmlRoot(Namespace = "urn:my-default-namespace:XSD:1")]
public class BankServiceResponse : ServiceResponseBase
{
// Determines if the class is being deserialized.
private bool _isDeserializing;
private ServiceResponseBase _serviceResponseBase;
// Constructor used by XmlSerializer.
// This is special because I require a non-null List<T> of items later on.
private BankServiceResponse()
{
this._isDeserializing = true;
this._serviceResponseBase = new ServiceResponseBase();
}
// Constructor used for unit testing
internal BankServiceResponse(bool isDeserializing = false)
{
this._isDeserializing = isDeserializing;
this._serviceResponseBase = new ServiceResponseBase();
}
public BankServiceResponse(Guid requestId, List<BankResponse> responses, Guid? asyncRequestId = null, Identifier name = null, Status status = null)
{
if (responses == null || responses.Count == 0)
throw new ArgumentNullException("The list cannot be null or empty", "responses");
this._serviceResponseBase = new ServiceResponseBase(requestId, asyncRequestId, name, status);
this._responses = responses;
}
[XmlElement(Order = 1)]
public Status Status
{
get { return this._serviceResponseBase.Status; }
set { this._serviceResponseBase.Status = value; }
}
[XmlElement(Order = 2)]
public Guid RequestId
{
get { return this._serviceResponseBase.RequestId; }
set { this._serviceResponseBase.RequestId = value; }
}
[XmlElement(Order = 3)]
public Guid? AsyncRequestId
{
get { return this._serviceResponseBase.AsyncRequestId; }
set { this._serviceResponseBase.AsyncRequestId = value; }
}
[XmlIgnore]
public bool AsyncRequestIdSpecified
{
get { return this._serviceResponseBase.AsyncRequestIdSpecified; }
set { ; } // Must have this for XmlSerializer.
}
[XmlElement(Order = 4)]
public Identifer Name
{
get { return this._serviceResponseBase.Name; }
set { this._serviceResponseBase.Name; }
}
[XmlElement(Order = 5)]
public List<BankResponse> Responses
{
get { return this._responses; }
set
{
if (this._isDeserializing && this._responses != null && this._responses.Count > 0)
this._isDeserializing = false;
if (!this._isDeserializing && (value == null || value.Count == 0))
throw new ArgumentNullException("List cannot be null or empty.", "value");
this._responses = value;
}
}
private List<BankResponse> _responses;
}
So, while I have to create properties for all of the contained classes, I can delegate any custom logic I might have within the contained class(es) property setters/getters by simply using the contained class's properties when the leaf class's properties are accessed. Since there's no inheritance, I can decorate all the properties of the leaf class with the XmlElementAttribute attribute and use any ordering that I see fit.
UPDATE:
I came back to revisit this article because my design decisions about using class inheritance came back to bite me again. While my solution above does work, I'm using it, I really think that Nader's solution is the best and should be considered before the solution I presented. In fact, I'm +1'ing him today! I really like his answer, and if I ever have the opportunity to refactor my current project, I will definitely be separating the business object from the serialization logic for objects that would otherwise benefit greatly from inheritance in order to simplify the code and make it easier for others to use and understand.
Thanks for posting your response Nader, as I think many will find it very instructive and useful.
