Reflection: Browsing Code Information, Part 2
This chapter is from the book
This chapter is from the book
This chapter is from the book
One of the most prominent uses for reflection is serialization. Serialization is the process of turning an object into a stream of bytes that can be stored or transferred "elsewhere" and then used to reconstruct the object. You obviously cannot take an existing object and directly transfer it to another machine and expect it to work. Certain addresses and handles only correspond to memory on one machine or, for that matter, one process. In the past, MFC had a facility whereby an object could serialize and deserialize itself. The programmer decided which fields needed to be serialized through the code. Using reflection, it was possible to assist the programmer in serializing an object. The programmer no longer needed to depend on intimate knowledge of an object's format to serialize it. In addition, if the object changed, it would automatically serialize in a correct manner without the need to change the code that references it.
Although an object can be serialized in many formats, the predominant format used for objects that are to be passed between disparate systems is XML. (Remoting opens the possibility of transferring data as a binary stream, which is much faster, but it is not very portable or compatible with other machines.) XML is the format that this article will primarily focus attention toward in discussing serialization.
The .NET Framework uses two types of serialization (binary serialization and SoapRpc serialization used with ASP.NET and Web applications have been excluded): one used for remoting and one used for "manual" serialization directly to XML. Although both of these serialization methods use XML as the underlying format, remoting uses a specific dialect of XML known as Simple Object Access Protocol (SOAP).
Serialization Used in Remoting
To demonstrate SOAP serialization that is used in remoting, consider the following code in Listings 1–3.
Listing 1—Building a Vehicle List
public class VehicleBuilder
{
public Vehicle Vehicle(string licenseNumber)
{
if (licenseNumber == "0")
{
Vehicle v = new Car();
v.licenseNumber = licenseNumber;
v.make = DateTime.Now;
return v;
}
else if (licenseNumber == "1")
{
Vehicle v = new Bike();
v.licenseNumber = licenseNumber;
v.make = DateTime.Now;
return v;
}
else
{
return null;
}
}
}
This is simply a utility class that allows you to easily build a vehicle based on the license number passed in ("0" is for a car, "1" is for a bike). Listing 2 shows the actual implementation of vehicle, car, and bike.
Listing 2—Defining a Vehicle
[Serializable]
public abstract class Vehicle
{
public string licenseNumber;
public DateTime make;
}
[Serializable]
public class Car : Vehicle
{
}
[Serializable]
public class Bike : Vehicle
{
}
Notice that all that is needed to serialize the object is to indicate that the object is serializable with the [Serializable] attribute. Vehicle is abstract to show that the serialization services can deal with polymorphism. That is all you have to do. Listing 3 shows a possible test.
Listing 3—Testing Serialization
public class VehicleSerializationTest
{
public static int Main(string[] args)
{
VehicleBuilder vb = new VehicleBuilder();
Vehicle [] va = new Vehicle[2];
va[0] = vb.Vehicle("0");
va[1] = vb.Vehicle("1");
// Create an instance of the Soap Serializer class;
// specify the type of object to serialize.
Stream stream = File.Create("out.xml");
SoapFormatter soap = new SoapFormatter();
soap.Serialize(stream, va);
stream.Close();
stream = File.OpenRead("out.xml");
va = (Vehicle [])soap.Deserialize(stream);
stream.Close();
Console.WriteLine("The first item is: {0}", va[0]);
Console.WriteLine("The second item is: {0}", va[1]);
return 0;
}
}
The first task is to build up an array of Vehicles. Because of the license number passed in the first (va[0]), Vehicle is a Car and the second is a Bike. Next, you open up (Create) a file called out.xml. Using the SoapFormatter class, you serialize the Vehicle Array out to the file. It is not important, but it is interesting to look at what the format of the output file looks like:
<SOAP-ENV:Envelope xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"xmlns:xsd="http://www.w3.org/2001/XMLSchema"
From this file, you can see that it created a Vehicle array with two Vehicles in it (arrayType="a1:Vehicle[2]"). SOAP creates two items referring to "#ref-3" and "#ref-4." Notice that "#ref-3" refers to a Car and "#ref-4" refers to a Bike (id="ref-3" and id="ref-4" respectively).
Go back to the sample in Listing 3. After the Vehicle array is serialized, then the file is immediately opened again and SoapFormatter deserializes the contents of the file, producing the Vehicle array again. Finally, each item is printed to the Console to verify that it was properly deserialized. The output looks like this:
The first item is: VehicleSerialization.Car The second item is: VehicleSerialization.Bike
You got back exactly what you put in.
The next example shows how the serializer can descend multiple levels of an object graph. Portions of the code will be shown in Listings 4 and 5. The full source for this sample is in Serialization\Multilevel\MultilevelSoap and is part of the solution in Serialization\Multilevel. You start out with a definition of the classes involved in Listing 4.
Listing 4—A Family Organization
[Serializable]
public class Person
{
public string name;
public double age;
}
[Serializable]
public class Marriage
{
public Person husband;
public Person wife;
}
[Serializable]
public class Family
{
public Marriage couple;
public Person [] children;
}
As you can see, this organization nests one class in another. This will test to see if you can maintain this structure when serializing and deserializing. Listing 5 shows the test used to exercise this serialization.
Listing 5—Building and Testing a Family Organization
public static int Main(string[] args)
{
Person me = new Person();
me.age = 43;
me.name = "Kevin Burton";
Person wife = new Person();
wife.age = 50;
wife.name = "Becky Burton";
Marriage marriage = new Marriage();
marriage.husband = me;
marriage.wife = wife;
Family family = new Family();
family.couple = marriage;
family.children = new Person[4];
family.children[0] = new Person();
family.children[0].age = 20;
family.children[0].name = "Sarah";
family.children[1] = new Person();
family.children[1].age = 18;
family.children[1].name = "Ann Marie";
family.children[2] = new Person();
family.children[2].age = 17;
family.children[2].name = "Mike";
family.children[3] = new Person();
family.children[3].age = 13;
family.children[3].name = "Rose";
Stream stream = File.Create("out.xml");
SoapFormatter soap = new SoapFormatter();
soap.Serialize(stream, family);
stream.Close();
Console.WriteLine(family.couple.husband.name);
Console.WriteLine(family.couple.wife.name);
foreach(Person p in family.children)
{
Console.WriteLine(p.name);
}
Console.WriteLine("----------");
stream = File.OpenRead("out.xml");
soap = new SoapFormatter();
family = (Family)soap.Deserialize(stream);
stream.Close();
Console.WriteLine(family.couple.husband.name);
Console.WriteLine(family.couple.wife.name);
foreach(Person p in family.children)
{
Console.WriteLine(p.name);
}
return 0;
}
The serialized output looks like this:
SOAP-ENV:Envelope xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
The output from the program looks like this:
Kevin Burton Becky Burton Sarah Ann Marie Mike Rose ---------- Kevin Burton Becky Burton Sarah Ann Marie Mike Rose
As was expected, what you put in is what you got out.
XML Serialization
To demonstrate XML serialization, or more precisely, serialization using System.Xml.Serialization, you will use the same classes that you used in the previous section. To start out, you will look at how the Vehicle class is serialized. Compare this implementation with the implementation described in Listings 1–3. Portions of the code will be shown in Listings 6–8. The full source to this sample is in Serialization\Vehicle\VehicleXml and is part of the solution in Serialization\Vehicle.
You will begin by generating a Vehicle array. No changes are required to the VehicleBuilder class, so for this sample, assume that you are using the same code as in Listing 1. Listing 6 shows how little is changed from the previous section.
Listing 6—Vehicles Marked for Serialization
[XmlInclude(typeof(Car)), XmlInclude(typeof(Bike))]
public abstract class Vehicle
{
public string licenseNumber;
public DateTime make;
}
public class Car : Vehicle
{
}
public class Bike : Vehicle
{
}
If you did not want to serialize an abstract class, you could have just removed the [Serializable] attribute from the classes and that would be all that needed to be done. As it is, you need to add an [XmlInclude] attribute; otherwise, the XmlSerializer gets confused when trying to serialize a Vehicle array. Listing 7 shows the test program.
Listing 7—Testing Vehicle Serialization with XmlSerializer
public static int Main(string[] args)
{
VehicleBuilder vb = new VehicleBuilder();
Vehicle [] va = new Vehicle[2];
va[0] = vb.Vehicle("0");
va[1] = vb.Vehicle("1");
// Create an instance of the XmlSerializer class;
// specify the type of object to serialize.
XmlSerializer serializer =
new XmlSerializer(typeof(Vehicle[]));
StreamWriter streamWriter = new StreamWriter("out.xml");
serializer.Serialize(streamWriter, va);
streamWriter.Close();
StreamReader streamReader = new StreamReader("out.xml");
va = (Vehicle [])serializer.Deserialize(streamReader);
streamReader.Close();
Console.WriteLine("The first item is: {0}", va[0]);
Console.WriteLine("The second item is: {0}", va[1]);
return 0;
}
Just like before, you create a Vehicle array first. Then instead of using a SoapFormatter, you use XmlSerializer. The Vehicle array is serialized to "out.xml". This file now looks like this:
<?xml version="1.0" encoding="utf-8"?> <ArrayOfVehicle xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
This is a little more readable than the SOAP version, but the same information is contained here. It has been said that programmatic types dominate serialization in System.Runtime.Serialization, which is used by remoting. In other words, the serialization scheme used for System.Runtime.Serialization is more centered in CLR types than in XML types if a conflict exists. With serialization using System.Xml.Serialization, the focus is more on XML types. A little of that bias can be seen in the requirement to use [XmlInclude] when serializing an abstract class. More importantly, using the attributes associated with System.Xml.Serialization gives the programmer a great deal of flexibility in mapping between CLR types and XML types.
This mapping is achieved by using an optional tool called the XML Schema Design Tool (xsd.exe). If the preceding vehicle class was put into a library (VehicleLibrary.dll), then you could run xsd.exe on the library like this:
xsd VehicleLibrary.dll
This would generate a file 'schema0.xsd' that looks like this:
<?xml version="1.0" encoding="utf-8"?> <xs:schema elementFormDefault="qualified"
This schema can be used to validate the XML coming into the system and that was generated by XmlSerializer. Because the sample that is shown both generates the XML and generates the schema as well, it is difficult to show the significance.
Suppose that the XML document must have the tags "Car" and "Bike" changed to "Automobile" and "Bicycle." In that case, you would just need to modify the Vehicle class:
[XmlType("Automobile")]
public class Car : Vehicle
{
}
[XmlType("Bicycle")]
public class Bike : Vehicle
{
}
Now the XML that is generated by the XmlSerializer looks like this:
<?xml version="1.0" encoding="utf-8"?> <ArrayOfVehicle xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
If the xsd.exe tool were run on the same library, it would generate the following:
<?xml version="1.0" encoding="utf-8"?> <xs:schema elementFormDefault="qualified"
Now the schema will properly validate XML documents coming into the system, and you will generate proper XML documents. (If this requirement were imposed on you, it would be likely that the external source or destination of the document would already have a schema that you would use to generate classes). Notice that the class names (in other words, the CLR types) have not changed. The other XML attributes allow for similar flexibility in mapping CLR types to XML types.
Late Binding to an Object Using Reflection
When the compiler encounters a reference when building an assembly, it loads the referenced assemblies and uses them to resolve all of the unresolved types. However, if you don't have the reference at compile time, you will not be able to call methods. You have just seen that with the classes in System.Reflection, you can dissect any assembly and find the methods and types in the assembly. It turns out that System.Reflection provides a means whereby a call can be made to a method in an assembly that is not known at compile time. This process is known as late binding.
It might be difficult to understand the usefulness of late binding at first, but some applications are not possible without a means of late binding. What if you wanted to provide feedback to third-party software? You don't know and don't want to know what the company will be doing with this feedback, but you want to provide a means whereby you can effectively notify them. You don't know ahead of time how many notifications you will have to set up. One solution would be to have a certain directory in which the third-party software would reside. You indicate to the developers of this third-party software that if they want to receive this notification, they will have to provide a function OutputMessage that takes as a single argument a string that contains the notification message. Now all you have to do is scan the directory and find out which of the software modules (probably DLLs) have an OutputMessage method defined and call the method. Using reflection, Listing 8 shows how this might be done with the .NET Framework. The complete source for this sample is in the LateBinding directory. Also included are two subprojects that build the required "first.dll" and "second.dll." If you need to use a directory other than the two directories supplied, or if you don't want to use debug mode, you will need to change the source so that it can locate the two DLLs.
Listing 8—Late Binding Client
static void Main(string[] args)
{
Assembly first = Assembly.LoadFrom(@"..\..\First\bin\debug\first.dll");
Assembly second = Assembly.LoadFrom(@"..\..\Second\bin\debug\second.dll");
Assembly [] assemblies = new Assembly[] { first, second };
foreach(Assembly a in assemblies)
{
Type[] types = a.GetTypes();
foreach(Type t in types)
{
try
{
object o = Activator.CreateInstance(t);
MethodInfo m = t.GetMethod("OutputMessage");
m.Invoke(o, new object [] {"This is my message"});
}
catch(Exception e)
{
Console.WriteLine(e);
}
}
}
}
Notice that directory scanning code has not been included for clarity. An Assembly array is available that you can scan for the OutputMessage method. The key portion of the code is the three lines beginning with the Activator.CreateInstance call. The Activator.CreateInstance creates an instance of the type specified. Here that type will be the class containing the OutputMessage method. A search is performed for the OutputMessage method within the Type. The return for this scan is an instance of the MethodInfo class that corresponds to the OutputMessage method. The method is then called with a single string argument.
As you can imagine, late binding is pretty slow compared to compiled early bound code, but in some cases, the flexibility provided by late binding more than justifies the extra CPU cycles.