Develop Reference

Circular reference in dependency injection

I am creating a RESTful api in Net 5, according to the instructions I must create repositories and services that make use of them. The logic must be in the services.
The Services I have are:
SubGroupService
GroupsService
The problem I have is that I generated a circular reference since in GroupsService I need a method of SubGroupsService and SubGroupsService i need a method of GroupsService .
Injecting the GroupsService service into SubGroupsService there is no problem, but injecting SubGroupsService into GroupsService generates the circular reference.
Please can you tell me how to solve this type of problem, since I don't have much experience with dependency injection.
SubGroupService
public class SubGroupService: ISubGroupService
{
private readonly ISubGroupRepository _SubGroupRepository;
private readonly IGroupService _GroupService;
public SubGroupService(
ISubGroupRepository SubGroupRepository,
IGroupService GroupService
{
_SubGroupRepository = SubGroupRepository;
_GroupService = GroupService;
}
public async Task InsertSubGroupService(Subgroup subgroup)
{
var group = await _GroupService.GetGroupService(subgroup.idgroup);
if (group != null)
{
await _SubGroupRepository.InsertSubGroupRepository(subgroup);
}
else
{
throw new BusinessException("This group not exists");
}
}
public async Task<Subgroups> GetSubGroupService(int idgroup)
{
return await _SubGroupRepository.GetSubGroupRepository(idgroup);
}
}
Group Service
public class GroupService : IGroupService
{
private readonly ISubGroupService _SubGroupService;
private readonly IGroupRepository _GroupRepository;
public GroupService(
ISubGroupService SubGroupService,
IGroupRepository GroupRepository)
{
_SubGroupService = SubGroupService;
_GroupRepository = GroupRepository;
}
public async Task<bool> DeleteGroupService(int Idgroup)
{
var existsSubGroup = await _SubGroupRepository(Idgroup);
if(existsSubGroup == null)
{
return await _GroupRepository.DeleteGroupRepository(Idgroup);
}
}
public async Task<Groups> GetGroupService(int Idgroup)
{
return await _GroupRepository.GetGroupRepository(Idgroup);
}
}
Interfaces:
public interface IGroupService
{
Task<Groups> GetGroupsService(int Idgroup);
Task<bool> DeleteGroupService(int Idgroup);
}
public interface ISubGroupService
{
Task<Subgroups> GetSubGroupService(int idsubgrupo);
Task InsertSubgroupService(Subgroup subgroup);
}

You can't use constructor injection in that case. You can switch to property injection:
public class SubGroupService: ISubGroupService
{
private readonly ISubGroupRepository _SubGroupRepository;
public IGroupService GroupService { get; set; }
public SubGroupService(
ISubGroupRepository SubGroupRepository)
{
_SubGroupRepository = SubGroupRepository;
}
// methods of the class
}
public class GroupService : IGroupService
{
public ISubGroupService SubGroupService {get; set;}
private readonly IGroupRepository _GroupRepository;
public GroupService(
IGroupRepository GroupRepository)
{
_GroupRepository = GroupRepository;
}
// methods of the class
}
You'll have to create the objects like this:
IGroupRepository groupRepository = new GroupRepository();
IGroupService groupService = new GroupService(groupRepository);
ISubGroupService subGroupService = new SubGroupService(groupRepository);
groupService.SubGroupSerivce = subGroupService;
subGroupService.GroupService = groupService;
Of course, creation of the objects is now much more complicated. You might put the creation into a facotry method to avoid errors:
public (IGroupService,ISubGroupService) CreateGroupAndSubGroupService()
{
// code from above
}
And it is also advisable to add null checks, because someone might create the objects without initializing the service correctly.

How to enforce method execution order in c#

Long story short, i have following class:
public class FlowBasePipeline<T>
{
private List<StepBaseBusiness<T>> stepList = new List<StepBaseBusiness<T>>();
public void Push(StepBaseBusiness<T> step)
{
stepList.Add(step);
}
public void Trigger(T result)
{
foreach (var step in stepList )
{
result = step.Execute(result);
if (!result.IsSuccess)
{
break;
}
}
}
}
What I'm looking for is forcing programmer to call Push method in the first place and then give them access to Trigger method, in this case following scenario is not allowed
var pipeline=new FlowBasePipeline<MyStepResult>();
pipeline.Trigger()// Trigger method is not recognized
we should first call Push method
var pipeline=new FlowBasePipeline<MyStepResult>();
pipeline.Push(new MyStep()).Trigger()//Now Trigger is recognized
What I've done:
I applied explicit interface method implementation as follows to get it to work:
public interface IBasePipeline<T> where T:BaseResult,new()
{
void Trigger();
IBasePipeline<T> Push(StepBaseBusiness<T> step);
}
public class FlowBasePipeline<T>:IBasePipeline<T> where T:BaseResult,new()
{
private List<StepBaseBusiness<T>> stepList = new List<StepBaseBusiness<T>>();
public IBasePipeline<T> Push(StepBaseBusiness<T> step)
{
stepList.Add(step);
return this;
}
void IBasePipeline<T>.Trigger(T result)
{
foreach (var step in stepList )
{
result = step.Execute(result);
if (!result.IsSuccess)
{
break;
}
}
}
}
Now it works well and we don't have access to Trigger method before Push method, but from my prospective it's not a good way as we might need more level of orders and i don't know how it could be done in this way.
As i know, method chaining is one of the key rules of functional programming.
Is there any pattern or strategy to implement this kind of chaining?
Update:
we need to call push method multiple times
var pipeline=new FlowBasePipeline<MyStepResult>();
pipeline.Push(new MyStep1()).Push(new MyStep2()).Trigger();
After first push, push and trigger would be available.

One way to do this is to use interfaces to restrict access to specific methods by specifying an interface as the result.
public interface IStartCar
{
IDriveCar Start(string key);
}
public interface IDriveCar
{
IParkCar Drive(string address);
}
public interface IParkCar
{
IStopCar Park();
}
public interface IStopCar
{
IParkCar Drive(string address);
void Stop();
}
public class Car : IStartCar, IDriveCar, IParkCar, IStopCar
{
public IDriveCar Start(string key);
public IParkCar Drive(string address);
public IStopCar Park();
public IStopCar Park();
private Car() { }
public static IStartCar Get()
{
var result = new Car();
return result;
}
}
Now to get a Car you use the CarFactory method Get(), it returns a car, but you really only have access to the interface result. This pattern only allows developers to string specific methods together:
var car = Car.Get();
car.Start("key").Drive("address1").Park().Drive("address2").Park().Stop();

Example of approach I use for api's that inherently 'guide' the callers with a fluent syntax:
public class Pipeline
{
readonly List<Action> _steps = new List<Action>();
// only Push is available when Pipeline is initialized
public PipelineWithSteps Push(Action step)
{
_steps.Add(step);
// or cache this if you want 'Push' repeatable
return new PipelineWithSteps(this);
}
public class PipelineWithSteps
{
// not required but often the chained context wants/needs access to the first context
readonly Pipeline _context;
// api is public but ctor cannot be invoked by external caller
internal PipelineWithSteps(Pipeline context) => _context = context;
// now Trigger is available only after something was pushed
public PipelineWithSteps Trigger()
{
foreach(var step in _context._steps)
step();
Console.WriteLine();
return this;
}
// usually I don't repeat an initialization method;
// this could be done using the 'context'
// but would have to be refactored to return the existing 'PipelineWithSteps'
public PipelineWithSteps Push(Action step)
{
_context._steps.Add(step);
return this;
}
}
}
Usage:
var pipeline = new Pipeline();
pipeline.Push(() => Console.WriteLine("A"))
.Push(() => Console.WriteLine("B"))
.Trigger()
.Push(() => Console.WriteLine("C"))
.Trigger();
Output:
A
B
A
B
C

After hours ,i came up with this design:
public interface IBasePipelineRegister<T> where T:BaseResult
{
IStagePipeline<T> Push(StepBaseBusiness<T> step);
List<StepBaseBusiness<T>> Steps { get; set; }
}
public interface IBasePipelineTrigger<T> where T:BaseResult
{
void Trigger(T result);
}
public interface IStagePipeline<T>: IBasePipelineTrigger<T>,IBasePipelineRegister<T> where T:BaseResult
{
}
public class FlowBasePipeline<TResult> : IBasePipelineRegister<TResult> where TResult : BaseResult,new()
{
public List<StepBaseBusiness<TResult>> Steps { get ; set ; }
private IStagePipeline<TResult> _stagePipeline;
public BasePipeline()
{
this.Steps = new List<StepBaseBusiness<TResult>>();
this._stagePipeline = new StagePipeline<TResult>(this);
}
public IStagePipeline<TResult> Push(StepBaseBusiness<TResult> step)
{
Steps.Add(step);
return _stagePipeline;
}
}
As you see, BasePipeline just implements IBasePipelineRegister and Register method presents new StagePipeline class that is consist of current class plus trigger implementation.
public class StagePipeline<T>: IStagePipeline<T> where T:BaseResult
{
private readonly IBasePipelineRegister<T> pipelineRegister;
public List<StepBaseBusiness<T>> Steps { get; set; }
public StagePipeline(IBasePipelineRegister<T> pipelineRegister)
{
this.pipelineRegister = pipelineRegister;
Steps = pipelineRegister.Steps;
}
public IStagePipeline<T> Push(StepBaseBusiness<T> step)
{
return pipelineRegister.Push(step);
}
public void Trigger(T result)
{
foreach (var step in Steps)
{
result = step.Execute(result);
if (!result.IsSuccess)
{
break;
}
}
}
}
Now each method adds a new feature not replacing new one.
var pipeline=new FlowBasePipeline<MyStepResult>();
pipeline.Push(new MyStep1()).Push(new MyStep2()).Trigger();

Appropriate design pattern for the payment modules c#

As i am learning through design pattern concept and also wanted to implement the payment modules in my project using the proper design pattern. So for that I have created some sample code.
Currently I have two concrete implementation for the payment PayPal and Credit Card. But the concrete implementation will be added further on the project.
Payment Service
public interface IPaymentService
{
void MakePayment<T>(T type) where T : class;
}
Credit Card and Pay Pal Service
public class CreditCardPayment : IPaymentService
{
public void MakePayment<T>(T type) where T : class
{
var creditCardModel = (CreditCardModel)(object)type;
//Implementation CreditCardPayment
}
}
class PayPalPayment : IPaymentService
{
public void MakePayment<T>(T type) where T : class
{
var payPalModel = (PayPalModel)(object)type;
//Further Implementation will goes here
}
}
Client Code Implementation
var obj = GetPaymentOption(payType);
obj.MakePayment<PayPalModel>(payPalModel);
Get Payment Option
private static IPaymentService GetPaymentOption(PaymentType paymentType)
{
IPaymentService paymentService = null;
switch (paymentType)
{
case PaymentType.PayPalPayment:
paymentService = new PayPalPayment();
break;
case PaymentType.CreditCardPayment:
paymentService = new CreditCardPayment();
break;
default:
break;
}
return paymentService;
}
I thought of implementing this modules using strategy design pattern, and I got deviated from Strategy and ended up doing this way.
Is this a proper way for creating the payment modules. Is there a more better approach of solving this scenario. Is this a design pattern?
Edited:
Client Code:
static void Main(string[] args)
{
PaymentStrategy paymentStrategy = null;
paymentStrategy = new PaymentStrategy(GetPaymentOption((PaymentType)1));
paymentStrategy.Pay<PayPalModel>(new PayPalModel() { UserName = "", Password = "" });
paymentStrategy = new PaymentStrategy(GetPaymentOption((PaymentType)2));
paymentStrategy.Pay<CreditCardModel>(
new CreditCardModel()
{
CardHolderName = "Aakash"
});
Console.ReadLine();
}
Strategy:
public class PaymentStrategy
{
private readonly IPaymentService paymentService;
public PaymentStrategy(IPaymentService paymentService)
{
this.paymentService = paymentService;
}
public void Pay<T>(T type) where T : class
{
paymentService.MakePayment(type);
}
}
Does this update inlines with the Strategy Pattern?

One major drawback of using an abstract factory for this is the fact that it contains a switch case statement. That inherently means if you want to add a payment service, you have to update the code in the factory class. This is a violation of the Open-Closed Principal which states that entities should be open for extension but closed for modification.
Note that using an Enum to switch between payment providers is also problematic for the same reason. This means that the list of services would have to change every time a payment service is added or removed. Even worse, a payment service can be removed from the strategy, but still be an Enum symbol for it even though it isn't valid.
On the other hand, using a strategy pattern doesn't require a switch case statement. As a result, there are no changes to existing classes when you add or remove a payment service. This, and the fact that the number of payment options will likely be capped at a small double-digit number makes the strategy pattern a better fit for this scenario.
Interfaces
// Empty interface just to ensure that we get a compile
// error if we pass a model that does not belong to our
// payment system.
public interface IPaymentModel { }
public interface IPaymentService
{
void MakePayment<T>(T model) where T : IPaymentModel;
bool AppliesTo(Type provider);
}
public interface IPaymentStrategy
{
void MakePayment<T>(T model) where T : IPaymentModel;
}
Models
public class CreditCardModel : IPaymentModel
{
public string CardHolderName { get; set; }
public string CardNumber { get; set; }
public int ExpirtationMonth { get; set; }
public int ExpirationYear { get; set; }
}
public class PayPalModel : IPaymentModel
{
public string UserName { get; set; }
public string Password { get; set; }
}
Payment Service Abstraction
Here is an abstract class that is used to hide the ugly details of casting to the concrete model type from the IPaymentService implementations.
public abstract class PaymentService<TModel> : IPaymentService
where TModel : IPaymentModel
{
public virtual bool AppliesTo(Type provider)
{
return typeof(TModel).Equals(provider);
}
public void MakePayment<T>(T model) where T : IPaymentModel
{
MakePayment((TModel)(object)model);
}
protected abstract void MakePayment(TModel model);
}
Payment Service Implementations
public class CreditCardPayment : PaymentService<CreditCardModel>
{
protected override void MakePayment(CreditCardModel model)
{
//Implementation CreditCardPayment
}
}
public class PayPalPayment : PaymentService<PayPalModel>
{
protected override void MakePayment(PayPalModel model)
{
//Implementation PayPalPayment
}
}
Payment Strategy
Here is the class that ties it all together. Its main purpose is to provide the selection functionality of the payment service based on the type of model passed. But unlike other examples here, it loosely couples the IPaymentService implementations so they are not directly referenced here. This means without changing the design, payment providers can be added or removed.
public class PaymentStrategy : IPaymentStrategy
{
private readonly IEnumerable<IPaymentService> paymentServices;
public PaymentStrategy(IEnumerable<IPaymentService> paymentServices)
{
this.paymentServices = paymentServices ?? throw new ArgumentNullException(nameof(paymentServices));
}
public void MakePayment<T>(T model) where T : IPaymentModel
{
GetPaymentService(model).MakePayment(model);
}
private IPaymentService GetPaymentService<T>(T model) where T : IPaymentModel
{
var result = paymentServices.FirstOrDefault(p => p.AppliesTo(model.GetType()));
if (result == null)
{
throw new InvalidOperationException(
$"Payment service for {model.GetType().ToString()} not registered.");
}
return result;
}
}
Usage
// I am showing this in code, but you would normally
// do this with your DI container in your composition
// root, and the instance would be created by injecting
// it somewhere.
var paymentStrategy = new PaymentStrategy(
new IPaymentService[]
{
new CreditCardPayment(), // <-- inject any dependencies here
new PayPalPayment() // <-- inject any dependencies here
});
// Then once it is injected, you simply do this...
var cc = new CreditCardModel() { CardHolderName = "Bob" /* Set other properties... */ };
paymentStrategy.MakePayment(cc);
// Or this...
var pp = new PayPalModel() { UserName = "Bob" /* Set other properties... */ };
paymentStrategy.MakePayment(pp);
Additional References:
Dependency Injection Unity - Conditional Resolving
Factory method with DI and IoC

This is one approach you could take. There's not a lot to go on from your source, and I'd really reconsider having MakePayment a void instead of something like an IPayResult.
public interface IPayModel { } // Worth investigating into common shared methods and properties for this
public interface IPaymentService
{
void MakePayment(IPayModel payModel);
}
public interface IPaymentService<T> : IPaymentService where T : IPayModel
{
void MakePayment(T payModel); // Void here? Is the status of the payment saved on the concrete pay model? Why not an IPayResult?
}
public class CreditCardModel : IPayModel
{
public string CardHolderName { get; set; }
}
public class PayPalModel : IPayModel
{
public string UserName { get; set; }
public string Password { get; set; }
}
public class CreditCardPayment : IPaymentService<CreditCardModel>
{
public void MakePayment(CreditCardModel payModel)
{
//Implmentation CreditCardPayment
}
void IPaymentService.MakePayment(IPayModel payModel)
{
MakePayment(payModel as CreditCardModel);
}
}
public class PayPalPayment : IPaymentService<PayPalModel>
{
public void MakePayment(PayPalModel payModel)
{
//Implmentation PayPalPayment
}
void IPaymentService.MakePayment(IPayModel payModel)
{
MakePayment(payModel as PayPalModel);
}
}
public enum PaymentType
{
PayPalPayment = 1,
CreditCardPayment = 2
}
So following your implementation approach, it could look something like:
static class Program
{
static void Main(object[] args)
{
IPaymentService paymentStrategy = null;
paymentStrategy = GetPaymentOption((PaymentType)1);
paymentStrategy.MakePayment(new PayPalModel { UserName = "", Password = "" });
paymentStrategy = GetPaymentOption((PaymentType)2);
paymentStrategy.MakePayment(new CreditCardModel { CardHolderName = "Aakash" });
Console.ReadLine();
}
private static IPaymentService GetPaymentOption(PaymentType paymentType)
{
switch (paymentType)
{
case PaymentType.PayPalPayment:
return new PayPalPayment();
case PaymentType.CreditCardPayment:
return new CreditCardPayment();
default:
throw new NotSupportedException($"Payment Type '{paymentType.ToString()}' Not Supported");
}
}
}
I also think for a strategy/factory pattern approach, manually creating an IPayModel type doesn't make much sense. Therefore you could expand the IPaymentService as an IPayModel factory:
public interface IPaymentService
{
IPayModel CreatePayModel();
void MakePayment(IPayModel payModel);
}
public interface IPaymentService<T> : IPaymentService where T : IPayModel
{
new T CreatePayModel();
void MakePayment(T payModel);
}
public class CreditCardPayment : IPaymentService<CreditCardModel>
{
public CreditCardModel CreatePayModel()
{
return new CreditCardModel();
}
public void MakePayment(CreditCardModel payModel)
{
//Implmentation CreditCardPayment
}
IPayModel IPaymentService.CreatePayModel()
{
return CreatePayModel();
}
void IPaymentService.MakePayment(IPayModel payModel)
{
MakePayment(payModel as CreditCardModel);
}
}
Usage would then be:
IPaymentService paymentStrategy = null;
paymentStrategy = GetPaymentOption((PaymentType)1);
var payModel = (PayPalModel)paymentStrategy.CreatePayModel();
payModel.UserName = "";
payModel.Password = "";
paymentStrategy.MakePayment(payModel);

Your code is basically using the factory pattern. This is a good way to handle more than one method of payment
http://www.dotnettricks.com/learn/designpatterns/factory-method-design-pattern-dotnet

Constructor dependency injection via unity with parameters from HttpContext

We're using domain to customize how our application behaves. I'll illustrate it on example:
// default behavior
public class CoreService : IService {
public virtual string Hello { get { return "Hello"; } }
public virtual string FavouriteDrink { get { return "Water"; } }
}
// german.site.com
public class GermanService : CoreService {
public override string Hello { get { return "Gutten tag"; } }
public override string FavouriteDrink { get { return "Beer"; } }
}
// usa.site.com
public class UsaService : CoreService {
public override string FavouriteDrink { get { return "Cofee"; } }
}
Services are bootstrapped as follow:
var container = new UnityContainer();
container.RegisterType<IService, CoreService>();
container.RegisterType<IService, GermanService>("german.site.com");
container.RegisterType<IService, UsaService>("usa.site.com");
I use Unity to bootstrap mvc controllers. IE:
public class HomeController : Controller {
private IService m_Service;
// contructor dependency injection magic - this resolves into "CoreService"
public HomeController([Dependency]IService service) {
if (service == null) {
throw new ArgumentNullException("service");
}
m_Service = service;
}
}
Is there a way how to change unity resolution so it'll take domain into account ? Right now I ended up with
public class HomeController : Controller {
private IService m_Service;
// contructor dependency injection magic - a lot less magical
public HomeController() {
m_Service = DomainServiceLocator.Retrieve<IService>();
}
}
Support classes:
public static class DomainServiceLocator {
private static UnityContainerAdapter adapter;
public static T Retrieve<T>() {
string domain = HttpContext.Current.Request.Url.Host;
if (adapter.IsServiceRegistered(typeof(T), domain)) {
return adapter.Resolve<T>(domain);
}
return adapter.Resolve<T>();
}
}
public class QueryableContainerExtension : UnityContainerExtension {
private List<RegisterInstanceEventArgs> registeredInstances = new List<RegisterInstanceEventArgs>();
private List<RegisterEventArgs> registeredTypes = new List<RegisterEventArgs>();
protected override void Initialize() {
this.Context.Registering += (sender, e) => { this.registeredTypes.Add(e); };
this.Context.RegisteringInstance += (sender, e) => { this.registeredInstances.Add(e); };
}
public bool IsServiceRegistered(Type service, string name) {
return registeredTypes.FirstOrDefault(e => e.TypeFrom == service && e.Name == name) != null
|| registeredInstances.FirstOrDefault(e => e.RegisteredType == service && e.Name == name) != null;
}
}
public class UnityContainerAdapter {
private readonly QueryableContainerExtension queryableContainerExtension;
private readonly IUnityContainer unityContainer;
public UnityContainerAdapter()
: this(new UnityContainer()) {
}
public UnityContainerAdapter(IUnityContainer unityContainer) {
this.unityContainer = unityContainer;
// adding extensions to unity container
this.queryableContainerExtension = new QueryableContainerExtension();
unityContainer.AddExtension(this.queryableContainerExtension);
}
public T Resolve<T>(string name) {
return unityContainer.Resolve<T>(name);
}
public T Resolve<T>() {
return unityContainer.Resolve<T>();
}
public bool IsServiceRegistered(Type service, string name) {
return this.queryableContainerExtension.IsServiceRegistered(service, name);
}
}

I like to use an injection factory in these scenarios when resolving something at runtime. Essentially you're resolving your type via the domain name:
So in your composition root you could register like this:
container.RegisterType<Func<string, IService>>
(
new InjectionFactory(c => new Func<string, IService>(name => c.Resolve<IService>(name)))
);
Then in your HomeController you can inject the delegate
public class HomeController
{
private readonly Func<string,IService> _serviceFactory;
public HomeController(Func<string, IService> serviceFactory)
{
if(serviceFactory==null)
throw new ArgumentNullException("serviceFactory");
this._serviceFactory= serviceFactory;
}
public void DoSomethingWithTheService()
{
var domain = this.HttpContext.Uri.Host;
var service = this._serviceFactory(domain);
var greeting = service.Hello;
}
}
```
This is then still unit testable and you have not leaked the DI contain implementation outside of "composition root".
Also.. should CoreService be abstract to avoid direct instantiation of it?

Below is the solution I ended up with - it is based on #Spencer idea. I've created a factory, default implementation to the factory has a reference to DI container itself (IUnityContainer in my case), so it can perform the resolution based on domain once it is asked to. It is also more "modern friendly" since in current generation of ASP.NET (ASP.NET CORE) there is no such thing as magic singleton providing current HttpContext and DI is hard coded into the framework.
public interface IFactory<T>
{
T Retrieve(string domain);
}
internal sealed class Factory<T> : IFactory<T>
{
private readonly IUnityContainer _container;
public Factory(IUnityContainer container)
{
_container = container;
}
public T Resolve(string domain)
{
// this is actually more complex - we have chain inheritance here
// for simplicity assume service is either registered for given
// domain or it throws an error
return _container.Resolve<T>(domain);
}
}
// bootstrapper
var container = new UnityContainer();
container.RegisterType<IService, CoreService>();
container.RegisterType<IService, GermanService>("german.site.com");
container.RegisterType<IService, UsaService>("usa.site.com");
container.RegisterInstance<IFactory<IService>>(new Factory<IService>(container));
And the home controller looks like
public class HomeController : Controller {
private IFactory<IService> m_Factory;
public HomeController(IFactory<IService> factory) {
m_Factory = factory;
}
private void FooBar() {
var service = m_Factory.Retrieve(this.HttpContext.Uri.Host);
var hello = service.Hello;
}
}
Its also a worth mentioning that - as I'm lazy - I've build a system of decorative attributes like
[Domain("german.site.com")]
public class GermanService : IService { ... }
[DomainRoot]
public class CoreService : IService { ... }
[Domain("usa.site.com")]
public class UsaService : CoreService { ... }
So the bootstrapping is done automatically across all types in given assembly. But that part is a bit lengthy - if anyone is interested I can post it on github.

How to export parts from an object not instantiated by the MEF container

Introduction
Class SessionModel is a service locator providing several services (I am going to elaborate my system architecture in the future, but for now I need to do it that way).
Code
I edited the following code part to be a Short, Self Contained, Correct (Compilable), Example (SSCCE):
using System;
using System.ComponentModel.Composition;
using System.ComponentModel.Composition.Hosting;
namespace ConsoleApplication1
{
internal class Program
{
private static void Main(string[] args)
{
var sessionModel = new SessionModel(3);
// first case (see text down below):
var compositionContainer = new CompositionContainer();
// second case (see text down below):
//var typeCatalog = new TypeCatalog(typeof (SessionModel));
//var compositionContainer = new CompositionContainer(typeCatalog);
compositionContainer.ComposeExportedValue(sessionModel);
var someService = compositionContainer.GetExportedValue<ISomeService>();
someService.DoSomething();
}
}
public class SessionModel
{
private int AValue { get; set; }
[Export]
public ISomeService SomeService { get; private set; }
public SessionModel(int aValue)
{
AValue = aValue;
// of course, there is much more to do here in reality:
SomeService = new SomeService();
}
}
public interface ISomeService
{
void DoSomething();
}
public class SomeService : ISomeService
{
public void DoSomething()
{
Console.WriteLine("DoSomething called");
}
}
}
Problem
I would like MEF to consider the parts (i.e. SomeService) exported by the service locator when composing other parts, but unfortunately this does not work.
First Case
When I try to get the exported value for ISomeService there is a System.ComponentModel.Composition.ImportCardinalityMismatchException telling me there are no exports with this contract name and required type identity (ConsoleApplication1.ISomeService).
Second Case
If I create the CompositionContainer using the TypeCatalog the exception is slightly different. It is a System.ComponentModel.Composition.CompositionException telling me MEF doesn't find a way to create a ConsoleApplication1.SessionModel (which is right and the reason why I am doing it myself).
Additional Information
mefx says for both cases:
[Part] ConsoleApplication1.SessionModel from: DirectoryCatalog (Path=".")
[Export] ConsoleApplication1.SessionModel.SomeService (ContractName="ConsoleApplication1.ISomeService")
[Part] ConsoleApplication1.SessionModel from: AssemblyCatalog (Assembly="ConsoleApplication1, Version=1.0.0.0, Culture=neutral, PublicKeyToken=null")
[Export] ConsoleApplication1.SessionModel.SomeService (ContractName="ConsoleApplication1.ISomeService")
What do I have to do? Is this possible with MEF or do I have to use Unity or StructureMap, or something else? Can this be done implementing an ExportProvider?

OK, that's how I did it:
I implemented my own SessionModelExportProvider finding exports in my SessionModel (see code below). Class SessionModelExport is just for holding the export data and – instead of creating an instance of a service – returning the value of the property of the SessionModel.
public class SessionModelExportProvider : ExportProvider
{
private List<Export> Exports { get; set; }
public SessionModelExportProvider(SessionModel sessionModel)
{
// get all the properties of the session model having an Export attribute
var typeOfSessionModel = typeof (SessionModel);
PropertyInfo[] properties = typeOfSessionModel.GetProperties();
var propertiesHavingAnExportAttribute =
from p in properties
let exportAttributes = p.GetCustomAttributes(typeof (ExportAttribute), false)
where exportAttributes.Length > 0
select new
{
PropertyInfo = p,
ExportAttributes = exportAttributes
};
// creating Export objects for each export
var exports = new List<Export>();
foreach (var propertyHavingAnExportAttribute in propertiesHavingAnExportAttribute)
{
var propertyInfo = propertyHavingAnExportAttribute.PropertyInfo;
foreach (ExportAttribute exportAttribute in propertyHavingAnExportAttribute.ExportAttributes)
{
string contractName = exportAttribute.ContractName;
if (string.IsNullOrEmpty(contractName))
{
Type contractType = exportAttribute.ContractType ?? propertyInfo.PropertyType;
contractName = contractType.FullName;
}
var metadata = new Dictionary<string, object>
{
{CompositionConstants.ExportTypeIdentityMetadataName, contractName},
{CompositionConstants.PartCreationPolicyMetadataName, CreationPolicy.Shared}
};
var exportDefinition = new ExportDefinition(contractName, metadata);
var export = new SessionModelExport(sessionModel, propertyInfo, exportDefinition);
exports.Add(export);
}
}
Exports = exports;
}
protected override IEnumerable<Export> GetExportsCore(ImportDefinition definition,
AtomicComposition atomicComposition)
{
return Exports.Where(e => definition.IsConstraintSatisfiedBy(e.Definition));
}
}
public class SessionModelExport : Export
{
private readonly SessionModel sessionModel;
private readonly PropertyInfo propertyInfo;
private readonly ExportDefinition definition;
public SessionModelExport(SessionModel sessionModel, PropertyInfo propertyInfo, ExportDefinition definition)
{
this.sessionModel = sessionModel;
this.propertyInfo = propertyInfo;
this.definition = definition;
}
public override ExportDefinition Definition
{
get { return definition; }
}
protected override object GetExportedValueCore()
{
var value = propertyInfo.GetValue(sessionModel, null);
return value;
}
}

The problem is that the SomeService is an instance property. You could have several SessionModel objects in your system, and MEF would have no way of knowing which SessionModel is returning the ISomeService instance that is supposed to be matched to an import.
Instead, just make SessionModel a static class and SomeService a static property. Alternatively, make SessionModel a singleton. The SomeService property would still be static, but would export the service from the one-and-only instance of SessionModel.

using System;
using System.ComponentModel.Composition;
using System.ComponentModel.Composition.Hosting;
using System.ComponentModel.Composition.ReflectionModel;
using System.Reflection;
using System.Linq;
namespace ConsoleApplication1
{
internal class Program
{
private static void Main(string[] args)
{
var catalogs = new AggregateCatalog();
var catalog = new System.ComponentModel.Composition.Hosting.AssemblyCatalog(Assembly.GetExecutingAssembly());
catalogs.Catalogs.Add(catalog);
var sessionModel = new SessionModel(3);
var container = new CompositionContainer(catalog);
ISomeService someService = container.GetExportedValueOrDefault<ISomeService>(sessionModel.cname);
if (someService != null)
{
someService.DoSomething();
}
}
}
public class SessionModel
{
private int AValue { get; set; }
//[Import("One",typeof(ISomeService))]
//public ISomeService SomeService { get; private set; }
public SessionModel(int aValue)
{
AValue = aValue;
// of course, there is much more to do here in reality:
}
public string cname { get { return "One"; } }
}
public class SessionModel1
{
private int AValue { get; set; }
//[Import("Two",typeof(ISomeService))]
//public ISomeService SomeService { get; private set; }
public SessionModel1(int aValue)
{
AValue = aValue;
}
public string cname { get { return "Two"; } }
}
public interface ISomeService
{
void DoSomething();
}
[Export("One",typeof(ISomeService))]
public class SomeService : ISomeService
{
public SomeService()
{
Console.WriteLine("Some Service Called");
}
public void DoSomething()
{
Console.WriteLine("DoSomething called");
Console.ReadKey();
}
}
[Export("Two",typeof(ISomeService))]
public class SomeService1 : ISomeService
{
public SomeService1()
{
Console.WriteLine("Some Service1 Called");
}
public void DoSomething()
{
Console.WriteLine("DoSomething called 1");
Console.ReadKey();
}
}
}

First case: By passing sessionModel to ComposeExportedValue you add a part of type SessionModel and not of ISomeService. To make this case work you nee to pass the service to ComposeExportedValue.
compositionContainer.ComposeExportedValue(sessionModel.SomeService);
Second case: In this case you leave the creation of parts to the container. The container can create new parts if there is either a parameter-less constructor or a constructor with parameters decorated with the ImportingConstructorAttribute. This most probably means that you will need to change your design a bit.
Personally I would go with the first case, but try to keep this to a minimum. After all the normal (and suggested) usage of MEF is letting the container create and handle parts.