Serialization

Serialization is the process of persisting the state of an object to a file or memory buffer. The persisted data contains all the necessary information you need to reconstruct (deserialize) the state of the object.

Serializable marks a class serializable.

[Serializable]
public class UserPrefs
{
  public string WindowColor;
  public int FontSize;

  [NotSerialized]
  private int passcode;
}

Example using BinaryFormatter to write a Serializable object to a file stream.

UserPrefs userData = new UserPrefs { WindowColor = "Blue", FontSize = 24 };
BinaryFormatter bf = new BinaryFormatter();
using (Stream fs = new File.OpenWrite("user.dat"))
{
  bf.Serialize(fs, userData); 
}

All classes in a hierarchy must be marked as serializable, otherwise using types like BinaryFormatter or SoapFormatter will throw a SerializationException at runtime.

Object Graphs

The CLR accounts for all related objects as well to ensure that the data is persisted correctly. It does this by building an object graph.

(3: Car) -> (2: Radio)  
(1: SpecialCar) -> (3: Car)  
(1: SpecialCar) -> (2: Radio)  

This is one way of representing a graph with three nodes, where: "Car" depends on "Radio", and "SpecialCar" depends on both "Car" and "Radio".

The CLR may represent this as such (specifics not certain from source material):

[Car 3, ref 2], [Radio 2], [SpecialCar 1, ref 3, ref 2] 

Serialization Formatters

.NET provides some types to do serialization.

• BinaryFormatter in System.Runtime.Serialization.Formatters.Binary
• SoapFormatter in System.Runtime.Serialization.Formatters.Soap
• XmlSerializer in System.Xml.Serialization

The first two serialize all fields of an object. The XML formatter, on the other hand, only serializes public fields.

Both the binary and SOAP formatters implement the IFormatter and IRemotingFormatter interfaces.

public interface IFormatter
{
  SerializationBinder Binder { get; set; }
  StreamingContext Context { get; set; }
  ISurrogateSelector SurrogateSelector { get; set; }
  object Deserialize(Stream serializationStream);
  void Serialize(Stream serializationStream, object graph);
}

public interface IRemotingFormatter : IFormatter
{
  object Deserialize(Stream serializationStream, HeaderHandler handler);
  void Serialize(Stream serializationStream, object graph, Header[] headers);
}

Collections

Most types in the System.Collections namespace are serializable.

Custom Serialization

The System.Runtime.Serialization namespace provides several types for this.

• ISerializable - You can implement this on a [Serializable] type to control its serialization.
• ObjectIDGenerator - Generates IDs for members in an object graph.
• [OnDeserialized] - Mark method to be called after deserialization.
• [OnDeserializing] - Mark method to be called before deserialization.
• [OnSerialized] - Mark method to be called on after serialization.
• [OnSerializing] - Mark method to be called on before serialization.
• [OptionalField] - Mark field that can be missing. Used for version tolerance, where if a field is missing in a new version of the type it can be safely ignored.
• SerializationInfo - This class is essentially a dictionary, or property bag, holding name-value pairs of the type's state data to be serialized.

When the BinaryFormatter serializes an object, the following information is saved:

• The fully qualified name of the objects in the graph (e.g. MyApp.MyClass).
• The name of the assembly defining the object graph (e.g. MyApp.exe).
• An instance of the SerializationInfo class that contains all stateful data maintained by the members in the object graph.

SoapFormatter follows a similar process, and XmlSerializer does not persist a type's fully qualified name or assembly name for portability.

Formatters also analyze the members in the object graph:

• A check is made to determine whether the object is marked with the [Serializable] attribute. If the object is not, a SerializationException is thrown.
• If the object is marked [Serializable], a check is made to determine whether the object implements the ISerializable interface. If this is the case, GetObjectData() is called on the object.
• If the object does not implement ISerializable, the default serialization process is used, serializing all fields not marked [NotSerialized].