We Keep Coding

Composite pattern of sealed class that has no interface

Let's say that I am using a library that I have no control over whatsoever. This library exposes service that requires argument of certain class. Class is marked as sealed and has no interface.
tl;dr: How can I reimplement sealed class as interface?
Code example:
using System;
namespace IDontHaveControlOverThis
{
// Note no interface and the class is being sealed
public sealed class ArgumentClass
{
public String AnyCall() => "ArgumentClass::AnyCall";
}
public sealed class ServiceClass
{
public String ServiceCall(ArgumentClass argument) => $"ServiceClass::ServiceCall({argument.AnyCall()})";
}
}
namespace MyCode
{
// Composite pattern, basically I need: "is a ArgumentClass"
// Obviously doesn't work - can't extend from sealed class
public class MyArgumentClass : IDontHaveControlOverThis.ArgumentClass
{
private IDontHaveControlOverThis.ArgumentClass arg = new IDontHaveControlOverThis.ArgumentClass();
public String AnyCall() => $"MyArgumentCLass::AnyCall({arg.AnyCall()})";
}
}
public class Program
{
public static void Main()
{
// I don't have control over this
IDontHaveControlOverThis.ServiceClass service = new IDontHaveControlOverThis.ServiceClass();
//This obviously works
IDontHaveControlOverThis.ArgumentClass arg = new IDontHaveControlOverThis.ArgumentClass();
Console.WriteLine($"Result: {service.ServiceCall(arg)}");
// How to make this work?
IDontHaveControlOverThis.ArgumentClass myArg = new MyCode.MyArgumentClass();
Console.WriteLine($"Result: {service.ServiceCall(myArg)}");
}
}

Based on the code sample you show, the answer is you can't. You need to be able to modify the behavior of IDontHaveControlOverThis.ArgumentClass, by setting a property, or creating a new instance with different constructor parameters in order to modify the servicecall. (It now always returns the same string, so the servicecall is always the same)
If you are able to modify the behavior of the ArgumentClass by setting properties.
You could create wrappers for the sealed classes in your own code, and use that throughout your codebase.
public class MyArgumentClass
{
// TODO: Set this to a useful value of ArgumentClass.
internal IDontHaveControlOverThis.ArgumentClass InnerArgumentClass { get; }
public virtual string AnyCall() => "???";
}
public class MyServiceClass
{
private IDontHaveControlOverThis.ServiceClass innerServiceClass
= new IDontHaveControlOverThis.ServiceClass();
public virtual string ServiceCall(MyArgumentClass argument)
{
return innerServiceClass.ServiceCall(argument.InnerArgumentClass);
}
}
or
public class MyArgumentClass
{
public virtual string AnyCall() => "???";
}
public class MyServiceClass
{
private IDontHaveControlOverThis.ServiceClass innerServiceClass
= new IDontHaveControlOverThis.ServiceClass();
public string ServiceCall(MyArgumentClass argument)
{
var serviceArgument = Convert(argument);
return innerServiceClass.ServiceCall(serviceArgument);
}
private IDontHaveControlOverThis.ArgumentClass Convert(MyArgumentClass argument)
{
// TODO: implement.
}
}

The compiler error message
Cannot implicitly convert type 'MyCode.MyArgumentClass' to 'IDontHaveControlOverThis.ArgumentClass'
note: emphasis mine
should give you a hint as to what you can do
public class MyArgumentClass {
private IDontHaveControlOverThis.ArgumentClass arg = new IDontHaveControlOverThis.ArgumentClass();
public String AnyCall() => $"MyArgumentCLass::AnyCall({arg.AnyCall()})";
public static implicit operator IDontHaveControlOverThis.ArgumentClass(MyArgumentClass source) {
return source.arg;
}
}
So now your "wrapper" exposes the 3rd party dependency as needed
IDontHaveControlOverThis.ArgumentClass myArg = new MyCode.MyArgumentClass();
or directly
var myArg = new MyCode.MyArgumentClass();
Console.WriteLine($"Result: {service.ServiceCall(myArg)}");
Reference User-defined conversion operators (C# reference)
Which can allow for abstracting your code
namespace MyCode {
public interface IMyService {
String ServiceCall(MyArgumentClass argument);
}
public class MyServiceClass : IMyService {
public string ServiceCall(MyArgumentClass argument) {
IDontHaveControlOverThis.ServiceClass service = new IDontHaveControlOverThis.ServiceClass();
return service.ServiceCall(argument);
}
}
}

Inherit in generic classes C#

My brain is gonna to explode. :) So I would like to get help from you.
Please, think about my question like about just programmer puzzle. (Actually. perhaps it is very easy question for you, but not for me.)
It is needed to create array of objects. For example List where T is class. (I will describe Class T below). Also it is needed create “container” that will contain this array and some methods for work with this array. For example Add(), Remove(int IndexToRemove).
Class T must have field "Container", this way each elements of our array would be able to know where is it contained and has access its container's fields and methods. Notice, that in this case Class T should have type parameter. Indeed, it is not known beforehand which container's type is used.
Let us denote this class container as A and class element (class T) as AUnit.
Code:
class Program
{
static void Main(string[] args)
{
A a = new A();
a.Add();
a.Units[0].SomeField +=100;
Console.ReadKey();
}
}
class A
{
public List<AUnit> Units;
public A()//ctor
{
Units = new List<AUnit>();
}
public void Add()
{
this.Units.Add(new AUnit(this));
}
}
class AUnit
{
public int SomeField;
public A Container;
public string Name { get; private set; }
public AUnit(A container)
{
this.SomeField = 43;
this.Container = container;
this.Name = "Default";
}
}
Public fields should be protected or private of course, but let think about this later.
You can ask “why we create public A Container field in AUnit”? We create field public string Name{get;private set;} (actually property but nevermind). And also we would like to be able to change value of this field for example method [Class AUnit] public bool Rename(string newName)();. The main idea of this method is changing Name field only that case if no one element in array (public List Units; ) has the same name like newName. But to achieve this, Rename method has to have access to all names that is currently used. And that is why we need Container field.
Code of extended version AUnit
class AUnit
{
public int SomeField;
public A Container;
public string Name { get; private set; }
public AUnit(A container)
{
this.SomeField = 43;
this.Container = container;
this.Name = "Default";
}
public bool Rename(String newName)
{
Boolean res = true;
foreach (AUnit unt in this.Container.Units)
{
if (unt.Name == newName)
{
res = false;
break;
}
}
if (res) this.Name = String.Copy(newName);
return res;
}
}
Ok. If you still read it let's continue. Now we need to create Class B and class BUnit which will be very similar like Class A and Class Aunit. And finally the main question of this puzzle is HOW WE CAN DO IT? Of course, I can CopyPaste and bit modify A and AUnit and create this code.
class B
{
public List<BUnit> Units; //Only Type Changing
public B()//ctor Name changing...
{
Units = new List<BUnit>();//Only Type Changing
}
public void Add()
{
this.Units.Add(new BUnit(this));//Only Type Changing
}
}
class BUnit
{
public int SomeField;
public B Container;//Only Type Changing
public string Name { get; private set; }
public A a; //NEW FIELD IS ADDED (just one)
public BUnit(B container) //Ctor Name and arguments type changing
{
this.SomeField = 43;
this.Container = container;
this.Name = "Default";
this.a=new A(); //New ROW (just one)
}
public bool Rename(String newName)
{
Boolean res = true;
foreach (BUnit unt in this.Container.Units) //Only Type Changing
{
if (unt.Name == newName)
{
res = false;
break;
}
}
if (res) this.Name = String.Copy(newName);
return res;
}
}
And I can to use this classes this way.
static void Main(string[] args)
{
B b = new B();
b.Add();
b.Units[0].a.Add();
b.Units[0].a.Units[0].SomeField += 100;
bool res= b.Units[0].a.Units[0].Rename("1");
res = b.Units[0].a.Units[0].Rename("1");
Console.ReadKey();
}
This construction is can be used to create “non-homogeneous trees”.
Help, I need somebody help, just no anybody…. [The Beatles]
I created B and BUnit using CopyPaste.
But how it can be done using “macro-definitions” or “Generic”, inherit or anything else in elegant style? (C# language)
I think that there is no reason to describe all my unsuccessful attempts and subquestions. Already topic is too long. : )
Thanks a lot if you still read it and understand what I would like to ask.

You need to implement a base type, lets call it UnitBase, with all common functionality. I'd structure your code the following way:
Create an interface for your container, this way you can change implementation to more performant solutions without modifying the elements you will be adding to the container.
public interface IContainer
{
Q Add<Q>() where Q : UnitBase, new();
IEnumerable<UnitBase> Units { get; }
}
Following the idea stated in 1, why not make the search logic belong to the container? It makes much more sense, as it will mostly depend on how the container is implemented:
public interface IContainer
{
Q Add<Q>() where Q : UnitBase, new();
IEnumerable<UnitBase> Units { get; }
bool Contains(string name);
}
A specific implementation of IContainer could be the following:
public class Container : IContainer
{
public Container()
{
list = new List<UnitBase>();
}
private List<UnitBase> list;
public Q Add<Q>() where Q: UnitBase, new()
{
var newItem = Activator.CreateInstance<Q>();
newItem.SetContainer(this);
list.Add(newItem);
return newItem;
}
public IEnumerable<UnitBase> Units => list.Select(i => i);
public bool Contains(string name) =>
Units.Any(unit => unit.Name == name);
}
Create a base class for your AUnit and BUnit types condensing all common functionality:
public abstract class UnitBase
{
protected UnitBase()
{
}
public IContainer Container { get; private set; }
public int SomeField;
public string Name { get; private set; }
public void SetContainer(IContainer container)
{
Container = container;
}
public bool Rename(String newName)
{
if (Container.Contains(newName))
return false;
this.Name = newName; //No need to use String.Copy
return true;
}
}
Implement your concrete types:
public class BUnit : UnitBase
{
public int SpecificBProperty { get; private set; }
public BUnit()
{
}
}
Shortcomings of this approach? Well, the container must be of type <UnitBase>, I've removed the generic type because it really wasn't doing much in this particular case as it would be invariant in the generic type.
Also, keep in mind that nothing in the type system avoids the following:
myContainer.Add<BUnit>();
myContainer.Add<AUnit>();
If having two different types in the same container is not an option then this whole set up kind of crumbles down. This issue was present in the previous solution too so its not something new, I simply forgot to point it out.

InBetween , I am very thankful to you for your advices. Actually I can't say that I understood your answer in full, but using your ideas I have done what I want.
Looks like my variant works well. However I would like to hear your (and everyone) opinions about code described below. The main goal of this structure is creating non-homogeneous trees. So could you estimate it from this side.
First of all. We need to create interfaces for both classes. We describe there all "cross-used" functions.
public interface IUnit<T>
{
string Name { get;}
void SetContainer(T t);
bool Rename(String newName);
}
public interface IContainer
{
bool IsNameBusy(String newName);
int Count { get; }
}
Next. Create Base for Unit Classes for future inheritance. We will use in this inheritors methods from Container Base so we need generic properties and IUnit interface.
class UnitBase<T> : IUnit<T> where T : IContainer
Unfortunately I don't know yet how to solve the problem with Constructor parameters. That is why I use method
SetContainer(T container).
Code:UnitBase
class UnitBase<T> : IUnit<T> where T : IContainer
{
protected T Container;
public string Name { get; private set; }
public UnitBase()
{
this.Name = "Default";
}
public void SetContainer(T container)
{
this.Container = container;
}
public bool Rename(String newName)
{
bool res = Container.IsNameBusy(newName);
if (!res) this.Name = String.Copy(newName);
return !res;
}
}
Next. Create ContainerBase
ContainerBase should:
1) has IContainer interface.
2)has information about what it will contain:
... where U : IUnit<C>, new()
3)and .... has information about what itself is. This information we need to pass as parameter to SetContainer() method.
Code ContainerBase:
class ContainerBase<U, C> : IContainer //U - Unit Class. C-Container Class
where U : IUnit<C>, new()
where C : ContainerBase<U, C>
{
protected List<U> Units;
public U this[int index] { get { return Units[index]; } }
public ContainerBase()//ctor
{
this.Units = new List<U>();
}
public void Add()
{
this.Units.Add(new U());
this.Units.Last().SetContainer(((C)this));//may be a bit strange but actualy this will have the same type as <C>
}
public bool IsNameBusy(String newName)
{
bool res = false;
foreach (var unt in this.Units)
{
if (unt.Name == newName)
{
res = true;
break;
}
}
return res;
}
public int Count { get { return this.Units.Count; } }
}
Cast ((TContainer)(this)) may be is a bit strange. But using ContainerBase we always should use NewInheritorContainer. So this cast is just do nothing…looks like...
Finally. This classes can be used like in this example.
class SheetContainer : ContainerBase<SheetUnit,SheetContainer> {public SheetContainer(){}}
class SheetUnit : UnitBase<SheetContainer>
{
public CellContainer Cells;
public PictureContainer Pictures;
public SheetUnit()
{
this.Cells = new CellContainer();
this.Pictures = new PictureContainer();
}
}
class CellContainer : ContainerBase<CellUnit, CellContainer> { public CellContainer() { } }
class CellUnit : UnitBase<CellContainer>
{
public string ValuePr;//Private Field
private const string ValuePrDefault = "Default";
public string Value//Property for Value
{
//All below are Just For Example.
get
{
return this.ValuePr;
}
set
{
if (String.IsNullOrEmpty(value))
{
this.ValuePr = ValuePrDefault;
}
else
{
this.ValuePr = String.Copy(value);
}
}
}
public CellUnit()
{
this.ValuePr = ValuePrDefault;
}
}
class PictureContainer : ContainerBase<PictureUnit, PictureContainer> { public PictureContainer() { } }
class PictureUnit : UnitBase<PictureContainer>
{
public int[,] Pixels{get;private set;}
public PictureUnit()
{
this.Pixels=new int[,]{{10,20,30},{11,12,13}};
}
public int GetSizeX()
{
return this.Pixels.GetLength(1);
}
public int GetSizeY()
{
return this.Pixels.GetLength(0);
}
public bool LoadFromFile(string path)
{
return false;
}
}
static void Main(string[] args)
{
SheetContainer Sheets = new SheetContainer();
Sheets.Add();
Sheets.Add();
Sheets.Add();
Sheets[0].Pictures.Add();
Sheets[1].Cells.Add();
Sheets[2].Pictures.Add();
Sheets[2].Cells.Add();
Sheets[2].Cells[0].Value = "FirstTest";
bool res= Sheets[0].Rename("First");//res=true
res=Sheets[2].Rename("First");//res =false
int res2 = Sheets.Count;
res2 = Sheets[2].Pictures[0].Pixels[1, 2];//13
res2 = Sheets[2].Pictures.Count;//1
res2 = Sheets[1].Pictures.Count;//0
res2 = Sheets[0].Pictures[0].GetSizeX();//3
Console.ReadKey();
}
Looks like it works like I want. But I didn’t test it full.
Let me say Thank you again, InBetween.

Unity.AutoRegistration ?? Convention Based Registration Unity

My code is as follows
IUnityContainer container = new UnityContainer();
container
.ConfigureAutoRegistration()
.LoadAssemblyFrom(typeof(Test).Assembly.Location)
.LoadAssemblyFrom(typeof(ITest).Assembly.Location)
.ApplyAutoRegistration();
This is my first question.
I'm not sure whether I have used the LoadAssemblyFrom method correctly here:
ITest test = container.Resolve<ITest>();
When I try to compile I get the exception "ResolutionFailedException".
What am I doing wrong?
Thanks for your time in advance.

It appears that what you are looking for is this:
container.ConfigureAutoRegistration()
.LoadAssemblyFrom(typeof(ITest).Assembly.Location)
.LoadAssemblyFrom(typeof(Test).Assembly.Location)
.Include(If.ImplementsITypeName, Then.Register())
.ApplyAutoRegistration();
This will tell Unity.AutoRegistration to register all types where there is an interface with the same name, prefixed with I.

Here is a complete working console example showing how to set Unity up for registration by convention, then transfer control into the dependency injection world. You will have to add the Unity NuGet package for this to work.
Tested with Unity v3.5 and VS 2012.
#region
using System;
using Microsoft.Practices.Unity;
#endregion
namespace Demo___Unity
{
internal class Program
{
private static void Main(string[] args)
{
using (var container = new UnityContainer())
{
// Manual method.
//container.RegisterType<IEntryPoint, EntryPoint>();
//container.RegisterType<IInjected, Injected>();
// Set up registration by convention.
// http://blogs.msdn.com/b/agile/archive/2013/03/12/unity-configuration-registration-by-convention.aspx
container.RegisterTypes(
AllClasses.FromAssembliesInBasePath(),
WithMappings.FromMatchingInterface,
WithName.Default,
WithLifetime.ContainerControlled);
var controller = container.Resolve<IEntryPoint>();
controller.Main();
}
}
}
public interface IEntryPoint
{
string Name { get; set; }
void Main();
}
public class EntryPoint : IEntryPoint
{
private readonly IInjected Injected;
public EntryPoint(IInjected injected)
{
Injected = injected;
}
public void Main()
{
Console.Write("Hello, world!\n");
Injected.SubMain();
Injected2.SubMain();
Console.Write("[any key to continue]");
Console.ReadKey();
}
// Demonstrates property injection.
[Dependency]
public IInjected Injected2 { get; set; }
public string Name { get; set; }
}
public interface IInjected
{
void SubMain();
}
public class Injected : IInjected
{
public void SubMain()
{
Console.Write("Hello, sub world!\n");
}
public string Name { get; set; }
}
}

C# add const field using attributes

We have class 'SomeClass':
namespace Namespace
{
class SomeClass
{
// something
}
}
And attribute 'SomeAttribute':
class SomeAttribute : System.Attribute { }
Task: add to all classes market by SomeAttribute 'public const string Type' field. Modified classes must be following:
class SomeClass
{
// something
public const string Type = #"Namespace.SomeClass";
}
UPD:
I'm using following approach for message transaction:
class Manager
{
// message has 3 parts:
// string message = String.Format("{0}{1}{2}",
// typeof(SomeClass).ToString(),
// splitter,
// Manager.Serialize(someClassObj)
// )
public static string GetType(string message) { /* some code */ }
public static string Serialize(SomeClass message) { /* XML serialization */ }
public static SomeClass Deserialize(string message) { /* deserialization */ }
}
class Logic
{
public void ProcessMessage(string message)
{
switch (Manager.GetType(message))
{
case SomeClass.Type:
{
SomeClass msg = Manager.Deserialize(message) as SomeClass;
// send message to binded objects
}
break;
case ClassInheritedFromSomeClass.Type:
{
// the same
}
break;
// etc.
}
}
}
UPD 2:
More about messages. At this time I'm using next approach:
public class BaseMessage
{
public const string Type = #"Messages.BaseMessage";
}
public class LoginMessage : BaseMessage
{
public new const string Type = #"Messages.Client.LoginMessage";
public string Nickname { get; set; }
public string Password { get; set; }
}
Conclusion
I think best case is to modify Manger like this:
class Manager
{
// create event table
public Action<BaseMessage> this[string eventName]
{
get
{
if (!m_eventTable.ContainsKey(eventName))
{
m_eventTable.Add(eventName, new Action<BaseMessage>(message => { }));
}
return m_eventTable[eventName];
}
set
{
m_eventTable[eventName] = value;
}
}
public void Send(BaseMessage message, string messageName)
{
if (m_eventTable.ContainsKey(messageName) && this[messageName].Method != null)
{
this[messageName].Invoke(message);
}
}
private Dictionary<string, Action<BaseMessage>> m_eventTable = new Dictionary<string, Action<BaseMessage>>();
}

Using switch with GetType is the wrong way to implement polymorphism, because it only checks the most-derived class (breaks extensibility).
In your particular case, where you want the Manager to be responsible for the behavior, you might use the dynamic keyword and overloaded methods. But this will again violate SOLID, because it isn't open for extension.
Instead of violating SOLID this way, try to find a way to use virtual methods to perform the type-specific action.

Getting only necessary plugins with MEF in .NET

I have IMessageSender interface.
using System.ComponentModel.Composition;
public interface IMessageSender
{
void Send(string message);
}
And I have two plugins that implements this interface. This is plugin.cs.
using System.ComponentModel.Composition;
using System.ComponentModel.Composition.Hosting;
using System.Reflection;
using System;
[Export(typeof(IMessageSender))]
public class EmailSender : IMessageSender
{
public void Send(string message)
{
Console.WriteLine(message);
}
}
and this is plugin2.cs
[Export(typeof(IMessageSender))]
public class EmailSender : IMessageSender
{
public void Send(string message)
{
Console.WriteLine(message + "!!!!");
}
}
And I have this code to run those plugins with MEF.
using System.ComponentModel.Composition;
using System.ComponentModel.Composition.Hosting;
using System.Reflection;
using System.Collections.Generic;
using System;
public class Program
{
[ImportMany]
public IEnumerable<IMessageSender> MessageSender { get; set; }
public static void Main(string[] args)
{
Program p = new Program();
p.Run();
foreach (var message in p.MessageSender) {
message.Send("hello, world");
}
}
public void Run()
{
Compose();
}
private void Compose()
{
var catalog = new AggregateCatalog();
catalog.Catalogs.Add(new DirectoryCatalog(#"./"));
var container = new CompositionContainer(catalog);
container.ComposeParts(this);
}
}
After compilation, I get what I want.
> mono program.exe
hello, world
hello, world!!!!
My question is how can I selectively run out of many plugins. This example just gets all the available plugins to run all of them, but what should I do when I just want to run first plugin or second plugin?
For example, can I run only plugin2.dll as follows?
public static void Main(string[] args)
{
Program p = new Program();
p.Run();
var message = messageSender.GetPlugin("plugin"); // ???
message.Send("hello, world");
}
SOLVED
Based on this site, and Matthew Abbott's answer. I could come up with this working code.
interface code (interface.cs)
using System.ComponentModel.Composition;
using System.ComponentModel.Composition.Hosting;
using System.Reflection;
using System;
public interface IMessageSender
{
void Send(string message);
}
public interface IMessageSenderMetadata
{
string Name {get; }
string Version {get; }
}
[MetadataAttribute]
[AttributeUsage(AttributeTargets.Class, AllowMultiple = false)]
public class MessageMetadataAttribute : ExportAttribute, IMessageSenderMetadata
{
public MessageMetadataAttribute( string name, string version)
: base(typeof(IMessageSender))
{
Name = name;
Version = version;
}
public string Name { get; set; }
public string Version { get; set; }
}
Plugin code (Plugin.cs ...)
using System.ComponentModel.Composition;
using System.ComponentModel.Composition.Hosting;
using System.Reflection;
using System;
[MessageMetadataAttribute("EmailSender1", "1.0.0.0")]
public class EmailSender : IMessageSender
{
public void Send(string message)
{
Console.WriteLine(message + "????");
}
}
Program.cs
using System.ComponentModel.Composition;
using System.ComponentModel.Composition.Hosting;
using System.Reflection;
using System.Collections.Generic;
using System;
using System.Linq;
public class Program
{
[ImportMany(typeof(IMessageSender), AllowRecomposition = true)]
public IEnumerable<Lazy<IMessageSender, IMessageSenderMetadata>> Senders { get; set; }
public static void Main(string[] args)
{
Program p = new Program();
p.Run();
var sender1 = p.GetMessageSender("EmailSender1","1.0.0.0");
sender1.Send("hello, world");
sender1 = p.GetMessageSender("EmailSender2","1.0.0.0");
sender1.Send("hello, world");
}
public void Run()
{
Compose();
}
public IMessageSender GetMessageSender(string name, string version)
{
return Senders
.Where(l => l.Metadata.Name.Equals(name) && l.Metadata.Version.Equals(version))
.Select(l => l.Value)
.FirstOrDefault();
}
private void Compose()
{
var catalog = new AggregateCatalog();
catalog.Catalogs.Add(new DirectoryCatalog(#"./"));
var container = new CompositionContainer(catalog);
container.ComposeParts(this);
}
}

MEF supports the exporting of custom metadata to accompany your exported types. What you need to do, is first define an interface that MEF will use to create a proxy object containing your metadata. In your example, you'll likely need a unique name for each export, so we could define:
public interface INameMetadata
{
string Name { get; }
}
What you would then need to do, is make sure you assign that metadata for each of your exports that require it:
[Export(typeof(IMessageSender)), ExportMetadata("Name", "EmailSender1")]
public class EmailSender : IMessageSender
{
public void Send(string message)
{
Console.WriteLine(message);
}
}
What MEF will do, is generate a project an implementation of your interface, INameMetadata using the value stored in the ExportMetadata("Name", "EmailSender1") atrribute.
After you've done that, you can do a little filtering, so redefine your [Import] to something like:
[ImportMany]
public IEnumerable<Lazy<IMessageSender, INameMetadata>> Senders { get; set; }
What MEF will create is an enumerable of Lazy<T, TMetadata> instances which support deferred instantiation of your instance type. We can query as:
public IMessageSender GetMessageSender(string name)
{
return Senders
.Where(l => l.Metadata.Name.Equals(name))
.Select(l => l.Value)
.FirstOrDefault();
}
Running this with an argument of "EmailSender1" for the name parameter will result in our instance of EmailSender being returned. The important thing to note is how we've selected a specific instance to use, based on querying the metadata associated with the type.
You can go one further, and you could amalgamate the Export and ExportMetadata attributes into a single attribute, such like:
[AttributeUsage(AttributeTargets.Class, AllowMultiple=false), MetadataAttribute]
public class ExportMessageSenderAttribute : ExportAttribute, INameMetadata
{
public ExportMessageSenderAttribute(string name)
: base(typeof(IMessageSender))
{
Name = name;
}
public string Name { get; private set; }
}
This allows us to use a single attribute to export a type, whilst still providing additional metadata:
[ExportMessageSender("EmailSender2")]
public class EmailSender : IMessageSender
{
public void Send(string message)
{
Console.WriteLine(message);
}
}
Obviously querying this way presents you with a design decision. Using Lazy<T, TMetadata> instances means that you'll be able to defer instantiation of the instance, but that does mean that only one instance can be created per lazy. The Silverlight variant of the MEF framework also supports the ExportFactory<T, TMetadata> type, which allows you to spin up new instances of T each time, whilist still providing you with the rich metadata mechanism.