Develop Reference

Elegantly handling base class dependencies

In the project I'm working on, I've added a base ViewModel class which contains some functionality and dependencies common to all ViewModels. It provides validation, messaging, dispatching and navigation services through the following properties:
IValidateProperties Validator { get; }
IMessenger Messenger { get; }
IDispatcherHelper DispatcherHelper { get; }
INavigationService Navigation { get; }
I use an IoC container to wire my dependencies, however I have a few options for how to handle these dependencies which are common to all of my ViewModels:
Inject them in the constructor. If I did this, then it would require adding these four arguments to the constructor of every single ViewModel and pass them to the base constructor. That's a lot of extra noise added to my code base, which I'd really rather avoid. Also, if I were to add another dependency at some point, that would require changing the constructor of every single ViewModel.
Use property injection. This is the approach I'm working with now. Unfortunately, it means not being able to access these properties until after the ViewModel has been constructed, resulting in the following workarounds:
private IValidateProperties _validator;
public IValidateProperties Validator
{
get => _validator;
set
{
_validator = value;
_validator.ValidationTarget = this;
}
}
private IMessenger _messenger;
public IMessenger Messenger
{
get => _messenger;
set
{
_messenger = value;
MessengerAttached();
}
}
protected virtual void MessengerAttached() { }
Make the properties static and inject them on app startup. This is easy for Messenger, DispatcherHelper and Navigation because they are used as singletons anyway. For Validator, I would need to add a static ValidatorFactory, and instantiate the Validator in the constructor using the factory. This way seems to be the cleanest way to do things, but I have this voice in the back of my head telling me that using statics like this is a bad idea.
I feel like option 1 is out of the question because of the large amount of noisy boilerplate code it would result in being added to my ViewModels. I'm still unsure whether 2 or 3 is the best way to go though. Is there a good reason that using statics is a bad idea in this case, or am I fretting over nothing?
Some people argue that statics lead to untestable code, but in this case it would actually make things easier to test. If I were to go with option 3, I could add the following class for all of my view model tests to inherit:
public abstract class ViewModelTestBase
{
protected readonly IValidateProperties ValidatorMock;
protected readonly IMessenger MessengerMock;
protected readonly IDispatcherHelper DispatcherHelperMock;
protected readonly INavigationService NavigationMock;
protected ViewModelTestBase()
{
ValidatorMock = Substitute.For<IValidateProperties>();
ViewModelBase.Validator = ValidatorMock;
MessengerMock = Substitute.For<IMessenger>();
ViewModelBase.Messenger = MessengerMock;
DispatcherHelperMock = Substitute.For<IDispatcherHelper>();
ViewModelBase.DispatcherHelper = DispatcherHelperMock;
NavigationMock = Substitute.For<INavigationService>();
ViewModelBase.Navigation = NavigationMock;
}
}
So, what concrete reasons are there for not going with approach #3? And if statics really are such a bad idea in this case, what concrete reasons are there for not going with approach #2 instead?

Pass parameter to a constructor in the ViewModel

I am building a WPF browser application with MVVM pattern.
I have a first page (ConsultInvoice) with a dataGrid. When I double click on one of the row I want to navigate to another page (EditInvoice) passing the selected row in argument to my constructor.
I know if I want do things properly I should use a dependency injection, but I don't really see how to use it here.
How can I simply pass this constructor?
ConsultInvoiceViewModel
private Invoice _selected;
public Invoice Selected
{
get
{
return _selected;
}
set
{
_selected = value;
OnPropertyChanged("Selected");
}
}
private void Edit()
{
EditInvoiceViewModel editInvoice = new EditInvoiceViewModel(Selected);
/* doing something here*/
}
public ICommand EditCommand
{
get
{
return editCommand ?? (editCommand = new RelayCommand(p => this.Edit(), p => this.CanEdit()));
}
}
EditInvoiceViewModel
public class EditInvoiceViewModel : ViewModelBase
{
public Context ctx = new Context();
Invoice invoice;
PreInvoice preInvoice;
#region properties
private ObservableCollection<PreInvoice> collection;
public ObservableCollection<PreInvoice> Collection
{
get
{
return collection;
}
set
{
collection = value;
OnPropertyChanged("Collection");
}
}
#endregion
public EditInvoiceViewModel(Invoice inv)
{
/* do stuff*/
}
}

Basically you should avoid passing such parameters into the ViewModels constructor, as wiring it with Inversion of Control/Dependency Injection becomes a pain. While you can use Abstract Factory pattern to resolve objects with runtime parameters, it's imho not suitable for ViewModels.
Instead I always suggest using a form of navigation pattern, similar to how Microsoft's Patterns & Practices team has done with Prism. There you have an INavigationAware interface which your ViewModels can implement. It has 2 methods, NavigateTo and NavigateFrom.
And there is a navigation service. The navigation service will switch the views and before switching calling NavigateFrom in the current ViewModel (if it implements it. One can use it to check if data is saved and if necessary cancel the navigation. After the new View has been loaded and the ViewModel assigned to it, call NavigateTo in the newly navigated ViewModel.
Here you'd pass the parameters required for the ViewModel, in your case invoiceId. Try avoid passing whole models or complex objects. Use the invoiceid to fetch the invoice data and to populate your editing ViewModel.
A basinc implementation from my former answer (can be found here):
public interface INavigationService
{
// T is whatever your base ViewModel class is called
void NavigateTo<T>() where T ViewModel;
void NavigateToNewWindow<T>();
void NavigateToNewWindow<T>(object parameter);
void NavigateTo<T>(object parameter);
}
public class NavigationService : INavigationService
{
private IUnityContainer container;
public NavigationService(IUnityContainer container)
{
this.container = container;
}
public void NavigateToWindow<T>(object parameter) where T : IView
{
// configure your IoC container to resolve a View for a given ViewModel
// i.e. container.Register<IPlotView, PlotWindow>(); in your
// composition root
IView view = container.Resolve<T>();
Window window = view as Window;
if(window!=null)
window.Show();
INavigationAware nav = view as INavigationAware;
if(nav!= null)
nav.NavigatedTo(parameter);
}
}
// IPlotView is an empty interface, only used to be able to resolve
// the PlotWindow w/o needing to reference to it's concrete implementation as
// calling navigationService.NavigateToWindow<PlotWindow>(userId); would violate
// MVVM pattern, where navigationService.NavigateToWindow<IPlotWindow>(userId); doesn't. There are also other ways involving strings or naming
// convention, but this is out of scope for this answer. IView would
// just implement "object DataContext { get; set; }" property, which is already
// implemented Control objects
public class PlotWindow : Window, IView, IPlotView
{
}
public class PlotViewModel : ViewModel, INotifyPropertyChanged, INavigationAware
{
private int plotId;
public void NavigatedTo(object parameter) where T : IView
{
if(!parameter is int)
return; // Wrong parameter type passed
this.plotId = (int)parameter;
Task.Start( () => {
// load the data
PlotData = LoadPlot(plotId);
});
}
private Plot plotData;
public Plot PlotData {
get { return plotData; }
set
{
if(plotData != value)
{
plotData = value;
OnPropertyChanged("PlotData");
}
}
}
}
An example of the INavigationAware interface used in Prism can be found on the projects github repository.
This makes it easy to pass parameter and async load your data (where there isn't any clean way to do this via constructor, as you can't await an async operation inside the constructor without locking, and doing this kind of things in the constructor is very discouraged).

Is this design pattern code smell? How to achieve this using DI?

I'm writing a WPF application using MVVM. My ViewModels are quite large and have a lot of logic associated with them (filtering, searching, writing to the database, etc), so I've decided to try to separate out the logic of the ViewModels to a "Presenter" class like is used in MVP.
So, my basic setup is this:
public class FooViewModel : ViewModelBase, IFooViewModel
{
private IFooPresenter presenter;
private ObservableCollection<FooModel> fooCollection;
public FooViewModel()
{
presenter = FooPresenter(this);
}
public ObservableCollection<FooModel> FooCollection
{
get { return fooCollection; }
set
{
fooCollection = value;
OnPropertyChanged("FooCollection");
}
}
public void FooCommandMethod(object obj)
{
presenter.DoStuff();
}
}
public class FooPresenter : IFooPresenter
{
private IFooViewModel viewModel;
public FooPresenter(IFooViewModel viewModel)
{
this.viewModel = viewModel;
}
public void DoStuff()
{
viewModel.FooCollection.Add(new FooModel());
//etc etc, make whatever ViewModel updates are needed
}
}
I feel like it is bad practice to have this circular dependency (View Model depends on Presenter and Presenter depends on View Model). These classes could be combined into one large ViewModel class, but I do like how clean this approach keeps my View Models, all that they do is hold commands that call presenter functions and hold the Model/collections of the Model. I also dislike the dependency of the ViewModel on the concrete implementation of the Presenter. One approach I have toyed with is using a Service Locator type class, so it would look like this:
public FooViewModel()
{
presenter = PresenterLocator.GetPresenter<IFooPresenter>(this);
}
What I would prefer, though, is to use Constructor Dependency Injection to inject the controller when I create the ViewModel. The problem with this is that this creates a circular dependency in the constructors of the ViewModels and Presenters, which causes my application to crash when I attempt to achieve this using Unity. It ends up looking like this:
public FooViewModel(IFooPresenter presenter)
{
this.presenter = presenterl
}
And
public FooPresenter(IFooViewModel viewModel(
{
this.viewModel = viewModel;
}
So, my concern is that my design approach is inherently flawed due to this. Nevertheless, I really like how clean it keeps my ViewModels and separates them from Business Logic. Is there a better way I could be designing this? Is there any way I can use DI to achieve this? Or by doing that am I essentially trying to force a DI container to act as a Service Locator?

First of all, I would not call this a "presenter". This introduces an unwanted confusion, in fact your presenter doesn't present anything, it is just an extracted bit of code from a large view model. Have you considered calling it just "a service"? A SearchService for example?
Another question is: does such service always depend on a view model? Or rather, could it depend on lower layers (unit of works/repos for example) or other services? Note that because your service depends on a view model and you pass a view model directly there, you loose a control of what happens to the view model inside a service. Your DoStuff method is a perfect example, it does something to a view model, alters its state. Instead, you could have
public class FooViewModel : ViewModelBase, IFooViewModel
{
private IFooService service;
private ObservableCollection<FooModel> fooCollection;
public FooViewModel()
{
service = FooService(this);
}
public void FooCommandMethod(object obj)
{
// the responsibility on consuming service outcome is still here!
this.FooCollection.Add( service.CreateNewModel() );
}
}
public class FooService : IFooService
{
// constructor parameter not needed now
public FooService()
{
this.viewModel = viewModel;
}
public FooModel CreateModel()
{
return ...;
}
}
If you still insist however on having a circular dependency, make it so that one of the two has a parameterless constructor and a property injector:
public class FooViewModel : IFooViewModel
{
private IFooService _service;
public FooViewModel( IFooService service )
{
this._service = service;
this._service.Model = this;
}
}
public class FooService : IFooService
{
public IFooViewModel Model { get; set; }
}
This way Unity asked for a IFooViewModel will resolve a parameterless IFooService and then execute the constructor that sets the cycle for both parties.

Polymorphism and Dependency injection

These days I'm facing this situation often and I'm looking for an elegant solution. I have :
public abstract class TypeA
{
public abstract void AbtractMethod(IDependency dependency);
}
public class TypeB : TypeA
{
public override void AbtractMethod(ISpecializedDependencyForB dependency) { }
}
public class TypeC : TypeA
{
public override void AbtractMethod(ISpecializedDependencyForC dependency) { }
}
public interface IDependency { }
public interface ISpecializedDependencyForB : IDependency { }
public interface ISpecializedDependencyForC : IDependency { }
My objective is to make things transparent in the client perspective and to consume this code like that :
TypeA myDomainObject = database.TypeARepository.GetById(id); // The important point here is that I don't know if the object is of TypeB or TypeC when I consume it.
IDependency dependency = ? // How do I get the right dependency
myDomainObject.AbtractMethod(dependency);
So the thing is that since I don't know the concrete type of the object, I can't inject the right dependency into it.
What I'm currently doing is that I create an abstract factory, to inject the right properties. I have two problems with that, the first one is that I would end up with a lot of factories. The second one is that it makes polymorphism useless since the client actually needs to care about "managing" the underlying type (I need to inject all the possible dependencies in the factory, and to instantiate the factory on the client code).
1) Therefore I was thinking of using property injection with unity, but I can't find out if it's possible to resolve the dependencies of an object, after it's been instanciated manually. Even with this approach I think I could still meet the same problem : I'm not sure if unity would check the actual type of the object and resolve the right dependency if a syntax like this existed :
unityContainer.Resolve<TypeA>(myDomainObject)
If not, I would need to know the type in advance and would be back to the same problem.
2) I have found this article mentionning that EF provides some mechanism for DI, but it seems that it is only meant to inject the framework services (PluralizationService, etc...). Otherwise it would have been a nice way to achieve that.
3) I could also not use DI in this case... It looks like by concept DI does not fit well with polymorphism. I'm not excited by this idea though.
I'd be happy to have a solution for the property injection I'm trying to achieve, or an idea of pattern I could use. However I really don't want to create a big infrastructure and obfuscate my code just for this purpose.
Note : I don't want to you use domain events in this case.
Thank you

TL;DR
Replace the IDependency parameter of the polymorphic AbstractMethod with an implementation-specific construction dependency parameter, which is injected by the IoC container, not by the consumer.
In more detail
The original class hierarchy will need to look like more like this for inheritance polymorphicism to work, as the superclass virtual method and subclass override methods must match signatures:
public abstract class TypeA // superclass
{
public abstract void AbtractMethod(IDependency dependency);
}
public class TypeB : TypeA // subclass 1
{
public override void AbtractMethod(IDependency dependency)
{
Contract.Requires(dependency is ISpecializedDependencyForB);
// ...
}
}
public class TypeC : TypeA // subclass 2
{
public override void AbtractMethod(IDependency dependency)
{
Contract.Requires(dependency is ISpecializedDependencyForC)
// ...
}
}
However, some things don't ring true with this design:
The LSP appears to be violated, since the although AbtractMethod() advertises that it accepts the base IDependency interface, the two subclasses actually depend on a specialized subclassed dependency.
It is also unusual, and arguably inconvenient, for a caller of these methods to build up the correct dependency and pass it to the method in order for it to be invoked correctly.
So, if possible, I would adopt a more conventional approach to the arrangement of dependencies, whereby the dependency is passed to the subclass constructor, and will be available to the polymorphic method when needed. This decouples the need to supply the appropriate IDependency to the method. Leave it to the IoC container to do the appropriate dependency resolution:
Use constructor injection to create the correct dependency into Classes TypeB and TypeC
If there is a secondary requirement to Expose an IDependency on the base class TypeA to consumers, then add an additional abstract property to TypeA of type IDependency (but this seems iffy)
As per Ewan's observation, the repository would need some kind of strategy pattern in order to serve up polymorphic domain entities (B or C). In which case, couple the repository to a factory to do exactly this. The concrete factory would need to be bound to the container in order to tap into Resolve().
So putting this all together, you might wind up with something like this:
using System;
using System.Diagnostics;
using Microsoft.Practices.Unity;
namespace SO29233419
{
public interface IDependency { }
public interface ISpecializedDependencyForB : IDependency { }
public interface ISpecializedDependencyForC : IDependency { }
public class ConcreteDependencyForB : ISpecializedDependencyForB {};
public class ConcreteDependencyForC : ISpecializedDependencyForC { };
public abstract class TypeA
{
// Your polymorphic method
public abstract void AbtractMethod();
// Only exposing this for the purpose of demonstration
public abstract IDependency Dependency { get; }
}
public class TypeB : TypeA
{
private readonly ISpecializedDependencyForB _dependency;
public TypeB(ISpecializedDependencyForB dependency)
{
_dependency = dependency;
}
public override void AbtractMethod()
{
// Do stuff with ISpecializedDependencyForB without leaking the dependency to the caller
}
// You hopefully won't need this prop
public override IDependency Dependency
{
get { return _dependency; }
}
}
public class TypeC : TypeA
{
private readonly ISpecializedDependencyForC _dependency;
public TypeC(ISpecializedDependencyForC dependency)
{
_dependency = dependency;
}
public override void AbtractMethod()
{
// Do stuff with ISpecializedDependencyForC without leaking the dependency to the caller
}
public override IDependency Dependency
{
get { return _dependency; }
}
}
public interface ITypeAFactory
{
TypeA CreateInstance(Type typeOfA);
}
public class ConcreteTypeAFactory : ITypeAFactory
{
private readonly IUnityContainer _container;
public ConcreteTypeAFactory(IUnityContainer container)
{
_container = container;
}
public TypeA CreateInstance(Type typeOfA)
{
return _container.Resolve(typeOfA) as TypeA;
}
}
public class TypeARepository
{
private readonly ITypeAFactory _factory;
public TypeARepository(ITypeAFactory factory)
{
_factory = factory;
}
public TypeA GetById(int id)
{
// As per Ewan, some kind of Strategy Pattern.
// e.g. fetching a record from a database and use a discriminating column etc.
return (id%2 == 0)
? _factory.CreateInstance(typeof (TypeB))
: _factory.CreateInstance(typeof (TypeC));
// Set the properties of the TypeA from the database after creation?
}
}
class Program
{
static void Main(string[] args)
{
// Unity Bootstrapping
var myContainer = new UnityContainer();
myContainer.RegisterType<ISpecializedDependencyForB, ConcreteDependencyForB>();
myContainer.RegisterType<ISpecializedDependencyForC, ConcreteDependencyForC>();
myContainer.RegisterType(typeof(TypeB));
myContainer.RegisterType(typeof(TypeC));
var factory = new ConcreteTypeAFactory(myContainer);
myContainer.RegisterInstance(factory);
myContainer.RegisterType<TypeARepository>(new InjectionFactory(c => new TypeARepository(factory)));
// And finally, your client code.
// Obviously your actual client would use Dependency Injection, not Service Location
var repository = myContainer.Resolve<TypeARepository>();
var evenNumberIsB = repository.GetById(100);
Debug.Assert(evenNumberIsB is TypeB);
Debug.Assert(evenNumberIsB.Dependency is ISpecializedDependencyForB);
var oddNumberIsC = repository.GetById(101);
Debug.Assert(oddNumberIsC is TypeC);
Debug.Assert(oddNumberIsC.Dependency is ISpecializedDependencyForC);
}
}
}

Could whatever it is that knows about the dependencies live behind an interface IDependencyProvider which has a function
IDependency GetDependency(Type type).
This could even just return an object and the class that realises the interface needs to know all the sub types and their associated dependencies.
AbstractMethod is then changed to:
void AbstractMethod(IDependencyProvider provider);
In your sub classes you then override this and call
var dependency = provider.GetDependency(this.GetType());
Your middle tier then knows nothing about the sub types or the sub dependencies.

It's an interesting problem, what I was thinking is that your repository knows about and creates the TypeB and TypeC classes and so you can add the correct dependency at that point
public class TypeARepository
{
private ISpecializedDependencyForB depB;
private ISpecializedDependencyForC depC;
public TypeARepository(ISpecializedDependencyForB depB, ISpecializedDependencyForC depC)
{
this.depB = depB;
this.depC = depC;
}
public TypeA GetById(string id)
{
if (id == "B")
{
return new TypeB(depB);
}
else
{
return new TypeC(depC);
}
}
}
The TypeB and TypeC would then implement their abstract methods with their private ref to the dependency rather than having it passed in in the method.
I come across this problem in various forms myself from time to time and it always seems to me that if there is that hard link between the types just having it setup via an injection config or the like is wrong. As it allows the installer to potentially set a bad config
This approach also allows you to inject your dependencies with unity

Thank you a lot for your interest in my question, I came up with a solution yesterday evening. The objective is to keep things transparent for the client and to take full advantage of polymorphism by syntaxes such as baseObjectReference.AbstractMethodCall().
I finally realized that I was able to achieve what I'm after by taking advantage of the static modifier and using it for DI purposes. So I have that :
public abstract class TypeA
{
public abstract void AbtractMethod();
}
public class TypeB : TypeA
{
private ISpecializedDependencyForB SpecializedDependencyForB
{
get
{
return GetSpecializedDependencyForB.CreateSpecializedDependencyForB();
}
}
public override void AbtractMethod() { // do stuff with dependency }
}
public static class GetSpecializedDependencyForB
{
public static ISpecializedDependencyForB DependencyForB
{
return CreateSpecializedDependencyForB();
}
public delegate ISpecializedDependencyForB CreateSpecializedDependencyForBDelegate();
public static CreateSpecializedDependencyForBDelegate CreateSpecializedDependencyForB;
}
And then, in my unity container I add this code :
public static void RegisterTypes(IUnityContainer container)
{
// .... registrations are here as usual
GetSpecializedDependencyForB.CreateSpecializedDependencyForB = CreateMyDomainService;
}
Having this method in the same unity config class :
private ISpecializedDependencyForB CreateMyDomainService()
{
return container.Value.Resolve<ISpecializedDependencyForB>();
}
And finally, I can simply use my object like this :
TypeA myDomainObject = database.TypeARepository.GetById(id);
myDomainObject.AbtractMethod();
And that's it !
So four things here :
The first one is that I inject the delegate that will create and instance of the service.
Then it is thread safe because static member is only written one time at the beginning of the application. All other accesses will be read. Moreover two threads won't share the same instance of the dependency since the delegate creates a new one all the time.
Also one interesting thing is that I can rely on my existing unity container configuration, no extra code is needed. It is important because my dependency may need other dependency to be constructed.
And finally the unity container is anyway also static, so there is no memory leak.
It's basically a manual and easy to set up "DI framework" sitting beside Unity.
And more importantly it works like a charm ! I'm finally satisfied with my design. I will only use this approach for polymorphic situations since injecting the right dependency in the method is easy for other situations. However it might be interesting to fully encapsulate the domain model using this approach.

Isolating a dependency for an instance (and that instances dependencies) when instance created through a Factory

EDIT: I've cleaned this question up significantly after solving my problem, including changing the title.
I have a MessageChannel interface which defines (unsurprisingly) a channel that my classes can use to push messages to the end user.
Normally this MessageChannel is a Singleton and is bound to a ViewModel that implements the MessageChannel interface. Essentially, there is a single location at the top of my application where messages to the user will be shown. So far its worked pretty well.
This MessageChannel is used in a lot of places, one of which is in some operation classes that I have setup.
I now have a need for a LOCAL MessageChannel, such messages being posted in some reduced scope get posted to that local MessageChannel and not the global one.
What this means is that I need to be able to create instances of a ViewModel (through a Factory), such that that particular instance has its own MessageChannel instance AND that MessageChannel instance is shared for all dependencies injected into that ViewModel (and their dependencies and so on).
Some classes to illustrate. I have simplified things somewhat, my messages are more than just strings:
using Ninject;
using Ninject.Extensions.Factory;
public interface MessageChannel
{
void PostMessage(string message);
}
public class MessageChannelViewModel : MessageChannel
{
public string Message { get; set; }
public void PostMessage(string message)
{
Message = message;
}
}
public interface Operation
{
void Do();
}
public interface OperationFactory
{
Operation Create();
}
public class DefaultOperation : Operation
{
public DefaultOperation(MessageChannel userMessages)
{
_UserMessages = userMessages;
}
private readonly MessageChannel _UserMessages;
public void Do()
{
// Do something.
_UserMessages.PostMessage("Success!");
}
}
public interface IsolatedViewModel
{
MessageChannelViewModel LocalMessages { get; }
}
public interface IsolatedViewModelFactory
{
IsolatedViewModel Create();
}
public class DefaultIsolatedViewModel : IsolatedViewModel
{
public IsolatedViewModel(MessageChannelViewModel localMessages, OperationFactory opFactory)
{
_OpFactory = opFactory;
LocalMessages = localMessages;
}
private readonly OperationFactory _OpFactory;
public MessageChannelViewModel LocalMessages { get; private set; }
}
public class Module : NinjectModule
{
public override void Load()
{
Bind<MessageChannel, MessageChannelViewModel>().To<MessageChannelViewModel>().InSingletonScope();
Bind<Operation>().To<DefaultOperation>();
Bind<OperationFactory>().ToFactory();
Bind<IsolatedViewModel>().To<DefaultIsolatedViewModel>();
Bind<IsolatedViewModelFactory>().ToFactory();
// Something to make it so the IsolatedViewModel DOESNT get the Singleton
// instance of the MessageChannelViewModel, and instead gets once of its own
// AND so the Operations created by the OperationFactory injected into the
// IsolatedViewModel get the SAME MessageChannel, so messages being posted
// from any place in the IsolatedViewModel's dependencies are shown only\
// locally.
}
}
I tried the NamedScope extension but I couldn't get it do what I wanted it to do.

I think you can try to use The Ninject Context Preservation Extension which adds support for recording (and making available to Contextual Binding rules) the context pertaining to factories that call the Kernel to Resolve Requests.
This enables you to add contextual conditions to your Bindings.

I ended up using a combination of Ninject.Extensions.ContextPreservation and Ninject.Extensions.NamedScope to accomplish what I wanted.
The completed example module looks like this:
public class Module : NinjectModule
{
public override void Load()
{
Bind<MessageChannel, MessageChannelViewModel>().To<MessageChannelViewModel>().InSingletonScope();
Bind<Operation>().To<DefaultOperation>();
Bind<OperationFactory>().ToFactory();
var uniqueName = "UNIQUE";
Bind<IsolatedViewModel>()
.To<DefaultIsolatedViewModel>()
.Named(uniqueName)
.DefinesNamedScope(uniqueName);
Bind<MessageChannel, MessageChannelViewModel>().To<MessageChannelViewModel>()
.WhenAnyAncestorNamed(uniqueName)
.InNamedScope(uniqueName);
Bind<IsolatedViewModelFactory>().ToFactory();
}
}
Theres two parts to it.
You need the ContextPreservation extension to choose the correct binding based on the context available at the time that you resolve the object instance. In this case I used a name as context, meaning that my special MessageChannel binding will be used when resolving the MessageChannel dependency for any dependencies required under the IsolatedViewModel.
I needed the NamedScope extension to ensure that only 1 instance of the MessageChannel was created under each IsolatedViewModel instance (i.e. the instance was shared for that IsolatedViewModel and all its dependencies).
Some other things to be aware of:
If you are using any ToMethod bindings and you use the Kernel inside the method, you'll need to make sure you use a ContextPreservingGet or you'll lose your context and the correct binding wont be selected.
You'll have to look very closely at your bindings and double check any Singleton bindings because if any of the dependencies of your isolated class are bound in Singleton scope, and they have dependencies on the MessageChannel (for example) its not going to work like you want it to. I had to remove a couple of Singleton scoped bindings as a result of this (for the better probably).