Wednesday, October 04, 2006

.NET deep clone - IsClone<T> using CloneFormatter

In some previous posts, I have written about how to implement an entity row state mechanism using .NET deep clone and an IsDirty method based on comparing two objects to see if they are clones of eachother. The IsClone<T> method was based on comparing the byte stream created by the BinaryFormatter, just as the Clone<T> method uses the BinaryFormatter to do deep cloning.

[UPDATE] A simple way to implement IsDirty when using IProperyNotificationChanged in data-binding enabled entity objects.

As I wrote in my last post, using the BinaryFormatter and the [NonSerialized] attribute in the IsClone<T> method is OK as long as you do not interfere with other stuff that use the BinaryFormatter, such as .NET remoting. The same applies to using the XmlFormatter, which will affect e.g. web-services. Thus, a separate formatter was needed to support the 'is clone' logic and fix the null/nothing string and the decimal problems. Enter the CloneFormatter based on the code in this article about Serialization Formatters.

I have made these changes to create the CloneFormatter:

private void WriteSerializableMembers(object obj, long objId)
System.Reflection.MemberInfo[] mi = FormatterServices.GetSerializableMembers(obj.GetType());
if (mi.Length > 0)
object[] od = FormatterServices.GetObjectData(obj, mi);
for (int i = 0; i < mi.Length; ++i)
System.Reflection.MemberInfo member = mi[i];
object data = od[i];
//KJELLSJ: do not serialize when marked with "CloneNonSerializedAttribute"
if (IsMarkedCloneNonSerialized(member) == true) continue;

if (member.MemberType == MemberTypes.Field)
FieldInfo fi = (FieldInfo)member;
if (data == null)
//KJELLSJ: ensure that null/nothing string gets "normalized"
if (fi.FieldType == typeof(string)) data = String.Empty;
//KJELLSJ: ensure that decimal gets "normalized"
if (fi.FieldType == typeof(decimal)) data = Convert.ToDecimal(Convert.ToDouble(data));
WriteMember(member.Name, data);

// Is the type attributed with [CloneNonSerialized]?
private bool IsMarkedCloneNonSerialized(MemberInfo mi)
object[] attributes = mi.GetCustomAttributes(typeof(CloneNonSerializedAttribute), false);
return attributes.Length > 0;

void ReadMember(long oid, MemberInfo mi, object o, SerializationInfo info)
//KJELLSJ: do not deserialize when marked with "CloneNonSerializedAttribute"
if (IsMarkedCloneNonSerialized(mi) == true) return;

// Read member name.
. . .

In addition, a new attribute was needed to be able to exclude class members from the 'is clone' logic:

[AttributeUsage(AttributeTargets.Field, AllowMultiple=false)]
public class CloneNonSerializedAttribute : System.Attribute
public CloneNonSerializedAttribute()
//only purpose is for marking fields as not-serialized by the CloneFormatter

Apply the [CloneNonSerialized] attribute to any class member that should not be part of the 'is clone' comparison.

The new IsClone<T> looks like this:

public static bool IsClone<T>(T sourceA, T sourceB)
if (sourceA == null sourceB == null)
return false;

MemoryStream streamA = new MemoryStream();
MemoryStream streamB = new MemoryStream();

IFormatter formatter = new CloneFormatter(); formatter.Serialize(streamA, sourceA);
//must create new formatter to reset the ObjectIDGenerator counter
formatter = new CloneFormatter();
formatter.Serialize(streamB, sourceB);

if (streamA.Length != streamB.Length) return false;

byte[] hashA = streamA.GetBuffer();
byte[] hashB = streamB.GetBuffer();

if (hashA.Length != hashB.Length) return false;

for (int i = 0; i < hashA.Length; i++)
if (hashA[i] != hashB[i]) return false;

//if here, objects are equal
return true;

Note that I have dropped the MD5 hashing in favor of plain byte-by-byte comparison as suggested by Nuri in a comment to the previous version. This should perform better, even if the for loop now gets longer that max 16 bytes.

Note also that the Clone<T> method still uses the BinaryFormatter to do the cloning. Use the CloneFormatter only for the IsClone<T> logic.

Note: serializing is not the most performant way to do cloning, read more at Anders Norås' blog.

No comments: