Dependency Injection: How to inject when using a multi-project solution

hope this question is not all too stupid, I'm trying to get a hold of more advanced programming principles and was thus trying to get used to Dependency Injection using Ninject.

So, my model is split up into several different .dll-projects. One project defines the model specifications (Interfaces) and a couple of others implement these interfaces. All model projects need to use some sort of database system, so they all need access to yet another .dll which implements all my database logic. It is important that all of them can access the same instance of my database object, though, so if wouldn't suffice to just create one instance for each model.

I'm not quite sure how to achieve this using dependency injection, though. My first thought was to create a seperate DI-project and bind all interfaces to their respective implementation, so the DI-project needed r开发者_开发百科eferences to all the other projects (model interfaces & implementations, database system etc.). Then again, the models would need access to the DI project since, for example, they'd need to request the database system from the DI System (Ninject). Of course this would create a circular reference (binding DI project to model and model to DI project), so it's not possible.

Long story short, I'd need a programming pattern that allows me to bind model interfaces to their implementations but that also allows the model implementations to request other dependencies from Ninject, e.g.

IModel1 -> Model1
IModel2 -> Model2 (different project)
IDatabase -> Database (different project)
Model1 -> request IDatabase -> get Database instance
Model2 -> request IDatabase -> get the same Database instance

Would be glad to get a couple of suggestions, at the moment I'm stuck and out of ideas ;) Thanks!

the models would need access to the DI project since, for example, they'd need to request the database system from the DI System (Ninject)

When you use dependency injection, the model shouldn't need to access the DI framework, since it is the DI framework that injects dependencies. The model objects should not be asking the DI container. When the your objects are asking the container for dependencies it is not called Dependency Injection, but Service Locator. Service Locator is an anti-pattern.

My first thought was to create a seperate DI-project

When you have a single application (e.g. a web app), the usual thing to do is completely configure the DI container in the startup project, as close as possible to the application’s entry point. This entry point where all object graphs are composed is called the Composition Root.

All model projects need to use some sort of database system, so they all need access to yet another .dll which implements all my database logic

Try making POCO (plain old CLR objects) model/entity objects, or at least, ensure that those objects don't need to reference any other project, which makes your architecture (and testing) much easier.

Use Ninject with the Register Resolve Release pattern from within a Composition Root.

The client application would use Ninject to inject the actual database and model implementations.

The client application therefore needs to reference the database, idatabase, model and imodel projects.

The idatabase and database projects need to reference the model project, as the methods will return model objects or collections of model objects. Have a look at the repository pattern.

Your model doesn't need to reference any of your projects.