Rich domain objects with DivineInject

DivineInject is a .net dependency injection framework, designed to be simple to use and easy to understand.

You can find the source for DivineInject on github. In this second part in the series we’ll look at creating rich domain objects, the first part covers getting started with Divine Inject.

Simple Domain Objects

As an example, imagine I have a simple shopping basket in my application. The shopping basket is encapsulated by a Basket object, for which the interface looks like this:

interface IBasket
{
    IList<IBasketItem> GetBasketContents();
    void AddItemToBasket(IBasketItem item);
}

The contents of the basket are actually backed by a service to provide persistence:

interface IBasketService
{
    IList<IBasketItem> GetBasketContents(Guid basketId);
    void AddItemToBasket(Guid basketId, IBasketItem item);
}

I can create a very simple implementation of my Basket:

class Basket : IBasket
{
    private readonly IBasketService _basketService;
    private readonly Guid _basketId = Guid.NewGuid();

    public Basket(IBasketService basketService)
    {
        _basketService = basketService;
    }

    public IList<IBasketItem> GetBasketContents()
    {
        return _basketService.GetBasketContents(_basketId);
    }

    public void AddItemToBasket(IBasketItem item)
    {
        _basketService.AddItemToBasket(_basketId, item);
    }
}

Since I’m using dependency injection, IBasketService sounds like a dependency, so how do I go about creating an instance of Basket? I don’t want to create it myself, I need the DI framework to create it for me, passing in dependencies.

I want to do something with how the Basket is created, so let’s start with a simple interface for creating baskets:

interface IBasketFactory
{
    IBasket Create();
}

When I’m creating a basket I don’t care about IBasketService or other dependencies; calling code just wants to be able to create a new, empty basket on demand. How would we implement this interface? Well, I could do the following – although I shouldn’t.

class BadBasketFactory : IBasketFactory
{
    public IBasket Create()
    {
        // DON'T DO THIS - just an example
        return new Basket(
            DivineInjector.Current.Get<IBasketService>());
    }
}

Now I’d never suggest actually doing this, explicitly calling the dependency injector. The last thing I want from my DI framework is to have references to it smeared all over the application. However, what this class does is basically what I want to happen.

DivineInject however can generate a class like this for you; this is configured at the same time you define the rest of your bindings:

DivineInjector.Current
    .Bind<IBasketFactory>().AsGeneratedFactoryFor<Basket>();

This generates an IBasketFactory implementation, which can create new IBasket implementations on demand (they will all actually be instances of Basket); all without having references to the DI framework smeared across my code. If I want to use the IBasketFactory, for example from my Session class, I declare it as a constructor arg the same as I would any other dependency:

public Session(IAuthenticationService authService, 
               IBasketFactory basketFactory)
{
    _authService = authService;
    _basketFactory = basketFactory;
}

The DI framework takes care of sorting out dependencies and I get a Session class with no references to DivineInject. When I need a new basket I just call _basketFactory.Create(). Since I have nice interfaces everywhere, everything is easy to mock so I can TDD everything.

Rich Domain Objects

Now what happens as my domain object becomes more complex? Say, for example, I want to be able to pass in some extra arguments to my constructor. Returning to our basket example: as well as starting a new, empty basket – isn’t there a possibility that I want to continue using an existing basket? E.g. in case of load balanced servers or fail-over. What do I do then?

I start by changing Basket, to allow me to pass in an existing basket id:

private readonly IBasketService _basketService;
private readonly Guid _basketId;

public Basket(IBasketService basketService)
{
    _basketService = basketService;
    _basketId = Guid.NewGuid();
}

public Basket(IBasketService basketService, Guid basketId)
{
    _basketService = basketService;
    _basketId = basketId;
}

I now have two constructors, one of which accepts a basket id. Since all Basket instances are created by an IBasketFactory, I need to change the factory interface, too:

interface IBasketFactory
{
    IBasket Create();
    IBasket UseExisting(Guid id);
}

I now have a new method on my IBasketFactory, if I was hand-coding the factory class I’d expect this second method to call the second constructor, passing in the basket id.

What do we need to tell DivineInject to make it generate this more complex IBasketFactory implementation? Nothing! That’s right, absolutely nothing – DivineInject will already generate a suitable IBasketFactory. Our original declaration above, is still sufficient:

DivineInjector.Current
    .Bind<IBasketFactory>().AsGeneratedFactoryFor<Basket>();

This generates an IBasketFactory implementation, returning a Basket instance for each method it finds on the interface. Since one of these methods takes a Guid, it tries to find a matching constructor which also takes a Guid, plus any dependencies it knows about. DivineInject can automatically wire up the right factory method to the right constructor, using the arguments it finds in each. Now, when a session wants to re-use an existing basket it just calls:

_basketFactory.UseExisting(existingBasketId)

This creates a new Basket instance, with dependencies wired up, passing in the basket id. Everything is still using interfaces so all your collaborations can be unit tested. Behind the scenes DivineInject generates the IL code to implement your factory interfaces, leaving you free to worry about your design.

By following this pattern we can create rich domain objects that include both state and dependencies: it becomes possible to create stateful objects that have behaviours (methods) that make sense in the domain. Successfully modelling your domain is critical to creating code that’s easy to understand and easy to change. DivineInject helps you model your domain better.

Your DI framework is killing your code

I read a really interesting post recently looking at the difference between typical OO code and a more functional style. There’s a lot to be said for the functional style of coding, even in OO languages like Java and C#. The biggest downside I find is always one of code organisation: OO gives you a discoverable way of organising large amounts of code. While in a functional style you might have less code, but it’s harder to organise it clearly.

It’s not Mike’s conclusion that I want to challenge, it’s his starting point: he starts out with what he describes as “typical object oriented C# code”. Unfortunately I think he’s bang on: even in this toy example, it is exactly like almost all so-called “object oriented” code I see in the wild. My issue with code like this is: it isn’t in the least bit object oriented. It’s procedural code haphazardly organised into classes. Just cos you’ve got classes, don’t make it OO.

What do I mean procedural code? The typical pattern I see is a codebase made up of two types of classes:

  1. Value objects, holding all the data with no business logic
    Extra credit here if it’s an object from your persistence layer, nhibernate or the like.
  2. Classes with one or two public functions –  they act on the value objects with no state of their own
    These are almost always “noun-verbers”

A noun-verber is the sure-fire smell of poor OO code: OrderProcessor, AccountManager, PriceCalculator. No, calling it an OrderService doesn’t make it any better. Its still a noun-verber you’re hiding by the meaningless word “service”. A noun-verber means its all function and no state, it’s acting on somebody else’s state. It’s almost certainly a sign of feature envy.

The other design smell with these noun-verbers is they’re almost always singletons. Oh you might not realise they’re singletons, because you’ve cleverly hidden that behind your dependency injection framework: but it’s still a singleton. If there’s no state on it, it might as well be a singleton. It’s a static method in all but name. Oh sure its more testable than if you’d actually used the word “static”. But it’s still a static method. If you’d not lied to yourself with your DI framework and written this as a static method, you’d recoil in horror. But because you’ve tarted it up and called it a “dependency”, you think it’s ok. Well it isn’t. It’s still crap code. What you’ve got is procedures, arbitrarily grouped into classes you laughably call “dependencies”. It sure as hell isn’t OO.

One of the properties of good OO design is that code that operates on data is located close to the data. The way the data is actually represented can be hidden behind behaviours. You focus on the behaviour of objects, not the representation of the data. This allows you to model the domain in terms of nouns that have behaviours. A good OO design should include classes that correspond to nouns in the domain, with behaviours that are verbs that act on those nouns: Order.SubmitForPicking(). UserAccount.UpdatePostalAddress(), Basket.CalculatePriceIncludingTaxes().

These are words that someone familiar with the domain but not software would still understand. Does your customer know what an OrderPriceStrategyFactory is for? No, then it’s not a real thing. Its some bullshit you made up.

The unloved value objects are, ironically, where the real answer is to the design problem. These are almost always actual nouns in the domain. They just lack any behaviours which would make them useful. Back to Mike’s example: he has a Customer class – its public interface is just an email address property, a classic value object: all state and no behaviour [if you want to play along at home, I’ve copied Mike’s code into a git repo; as well as my refactored version].

Customer sounds like a good noun in this domain. I bet the customer would know what a Customer was. If only there were some behaviours this domain object could have. What do Customers do in Mike’s world? Well, this example is all about creating and sending a report. A report is made for a single customer, so I think we could ask the Customer to create its own report. By moving the method from ReportBuilder onto Customer, it is right next to the data it operates on. Suddenly the public email property can be hidden – well this seems a useful change, if we change how we contact customers then only the customer needs to change, not also the ReportBuilder. It’s almost like a single change in the design should be contained within a single class. Maybe someone should write a principle about this single responsibility malarkey…

By following a pattern like this, moving methods from noun-verbers onto the nouns on which they operate, we end up with a clearer design. A Customer can CreateReport(), a Report can RenderAsEmail(), and an Email can Send(). In a domain like this, these seem like obvious nouns and obvious behaviours for these nouns to have. They are all meaningful outside of the made up world of software. The design models the domain and it should be clear how the domain model must change in response to each new requirement – since each will represent a change in our understanding of the domain.

So why is this pattern so uncommon? I blame the IoC frameworks. No seriously, they’re completely evil. The first thing I hit when refactoring Mike’s example, even using poor man’s DI, was that my domain objects needed dependencies. Because I’ve now moved the functionality to email a report from ReportingService onto the Report domain object, my domain object now needs to know about Emailer. In the original design it could be injected in, but with an object that’s new’d up, how can I inject the dependency? I either need to pass Emailer into my Report at construction time or on sending as email. When refactoring this I opted to pass in the dependency when it was used, but only because passing in at construction time is cumbersome without support.

Almost all DI frameworks make this a royal pain. If I want to inject dependencies into a class that also has state (like the details of the customer’s report), it’s basically impossible, so nobody does it. It’s better, simpler, quicker to just pull report creation onto a ReportBuilder and leave Customer alone. But it’s wrong. Customer deserves to have some functionality. He wants to be useful. He’s tired of just being a repository for values. If only there was a way of injecting dependencies into your nouns that wasn’t completely bullshit.

For example using NInject – pretty typical of DI frameworks – you can create a domain object that requires injected dependencies and state by string typing. Seriously? In the 21st century, you want me to abandon type safety, and put the names of parameters into strings. No. Just say no.

No wonder when faced with these compromises people settle for noun-verbers. The alternatives are absolutely horrific. The only DI framework I’ve seen which lets you do this properly is Guice‘s assisted injection. Everybody else, as far as I can see, is just plain wrong.

Is your DI framework killing your code? Almost certainly: if you’ve got value objects and noun-verbers, your design is rubbish. And it’s your DI framework’s fault for making a better design too hard.

Dependency injection with PostSharp

I don’t really like IoC containers. Or rather, I don’t like the crappy code people write when they’re given an IoC container. Before you know it you have NounVerbers everywhere, a million dependencies and no decent domain model. Dependencies should really be external to your application; everything outside of the core domain model that your application represents.

  • A web service? That’s a dependency
  • A database? Yup
  • A message queue? Definitely a dependency
  • A scheduler or thread pool? Yup
  • Any NounVerber (PriceCalculator, StockFetcher, BasketFactory, VatCalculator) no! Not a dependency. Stop it. They’re all part of your core business domain and are actually methods on a class. If you can’t write Price.Calculate() or Stock.Fetch() or new Basket() or Vat.Calculate() then fix your domain model first before you go hurting yourself with an IoC container

A while back I described a very simple, hand-rolled approach to dependency injection. But if we wave a little PostSharp magic we can improve on that basic idea. All the source code for this is available on github.

It works like this: if we have a dependency, say an AuthService, we declare an interface that business objects can implement to request that they have the dependency injected into them. In this case, IRequireAuthService.

class User : IRequireAuthService
{
  public IAuthService AuthService { set; private get; }

We create a DependencyInjector that can set these properties:

public void InjectDependencies(object instance)
{
  if (instance is IRequireAuthService)
    ((IRequireAuthService)instance).AuthService = AuthService;
    ...
}

This might not be the prettiest method – you’ll end up with an if…is IRequire… line for each dependency you can inject. But this provides a certain amount of friction. While it is easy to add new dependencies, developers are discouraged from doing it. This small amount of friction massively limits the unchecked growth of dependencies so prevalent with IoC containers. This friction is why I prefer the hand-rolled approach to off-the-shelf IoC containers.

So how do we trigger the dependency injector to do what it has to do? This is where some PostSharp magic comes in. We declare an attribute to use on the constructor:

  [InjectDependencies]
  public User(string id)

Via the magic of PostSharp aspect weaving this attribute causes some code to be executed before the constructor. This attribute is simply defined as:

public sealed override void OnEntry(MethodExecutionArgs args)
{
  DependencyInjector.CurrentInjector.InjectDependencies(args.Instance);
  base.OnEntry(args);
}

And that’s it – PostSharp weaves this method before each constructor with the [InjectDependencies] attribute. We get the current dependency injector and pass in the object instance (i.e. the newly created User instance) to have dependencies injected into it. Just like that we have a very simple dependency injector. Even better all this aspect weaving magic is available with the express (free!) edition of PostSharp.

Taking it Further

There are a couple of obvious extensions to this. You can create a TestDependencyInjector so that your unit tests can provide their own (mock) implementations of dependencies. This can also include standard (stub) implementations of some dependencies. E.g. a dependency that manages cross-thread scheduling can be replaced by an immediate (synchronous) implementation for unit tests to ensure that unit tests are single-threaded and repeatable.

Secondly, the DependencyInjector uses a ThreadLocal to store the current dependency injector. If you use background threads and want dependency injection to work there, you need a way of pushing the dependency injector onto the background thread. This generally means wrapping thread spawning code (which will itself be a dependency). You’ll want to wrap any threading code anyway to make it unit-testable.

Compile Time Checks

Finally, the most common failure mode we encountered with this was people forgetting to put [InjectDependencies] on the constructor. This means you get nulls at runtime, instead of dependencies. With a bit more PostSharp magic (this brand of magic requires the paid-for version, though) we can stop that, too. First, we change each IRequire to use a new attribute that indicates it manages injection of a dependency:

[Dependency]
public interface IRequireAuthService
{
  IAuthService AuthService { set; }
}

We configure this attribute to be inherited to all implementation classes – so all business objects that require auth service get the behaviour – then we define a compile time check to verify that the constructors have [InjectDependencies] defined:

public override bool CompileTimeValidate(System.Reflection.MethodBase method)
{
  if (!method.CustomAttributes.Any(a => a.AttributeType == typeof(InjectDependenciesAttribute)))
  {
    Message.Write(SeverityType.Error, "InjectDependences", "No [InjectDependencies] declared on " + method.DeclaringType.FullName + "." + method.Name, method);
    return false;
  }
  return base.CompileTimeValidate(method);
}

This compile time check now makes the build fail if I ever declare a class IRequireAuthService without adding [InjectDependencies] onto the class’ constructor.

Simple, hand-rolled dependency injection with compile time validation thanks to PostSharp!

 

I can haz dependencies?

I hate IoC containers. Spring? Evil. Guice? The devil’s own work. Why? Because it leads to such slack, lazy, thoughtless programming.

Why the hate?

Ok, perhaps I better explain myself a bit. IoC is a great idea. What annoys me, is the way IoC frameworks end up getting used by normal people. I’ve ranted previously about how IoC containers lead us to implement anaemic domain models. The trouble is, once you have a hammer, everything starts to look like a nail. Especially those pesky fingers. Once you have a dependency injection framework, everything starts to look like a dependency that needs to be injected. Need to implement some business logic? First create a new class, test drive it, then make it injectable, inject it into the class where the calling code needs it, test driving it natch, then bingo – you just hit yourself on the finger.

Now I’ve got two classes, basically closely coupled, but the IoC container hides that fact from me. I see a nice, clean interface being injected in. Aren’t I a good little OO developer? No, you’re stupid and you’re lazy.

Before you know it, your class has a dozen or more dependencies, each of which have a dozen dependencies, each of which have a dozen dependencies, each of which… you get the picture. You’ve managed to build a rats nest of a dependency graph, little by little. What you’ve TDD’d isn’t a design. The technical name for it is The Big Ball of Mud.

An Alternative

Instead, I think dependency injection works best at application seams, at architectural boundaries. Say, for example, you’re building a web app. You’ve created a TradeEntryController that allows users to, well, enter trades. The TradeEntryController naturally has loads of dependencies on the rest of the system. It needs to fetch valid assets to invest in and prices, it needs to know what your balance is so you can’t buy more shares than money in the bank etc etc. A perfect example where life without an IoC container could become really cumbersome.

But, I don’t think you need one. I think what your controller needs is a few, specific dependencies – that define the architectural boundary the controller lives within. Above the controller is a HTTP request, a session and all that blah blah. Within it, is business logic. Below it is the database. So, the dependencies we inject should represent only the architectural context in which the controller operates. For the most part, this will be common to all my controllers – not just trade entry. Controllers for managing balances, lists of assets, user accounts – these all depend on knowing stuff about their session, and to be able to talk to the next layer down: the database (or in an n-tier setup, perhaps some web services).

So, why not just inject those dependencies?

public class TradeEntryController {
    public void setSessionManager(ISessionManager sessionManager) { ... }
    public void setTradeDatabase(ITradeDatabase tradeDatabase) { ... }
    public void setAccountDatabase(IAccountDatabase accountDatabase) { ... }
    public void setAssetDatabase(IAssetDatabase assetDatabase) { ... }
}

Then in my controller, I can fetch user information from the SessionManager; I can get the list of assets from the AssetDatabase; I can check the user’s balance via the AccountDatabase; and I can record the trade via the TradeDatabase. So far, so much the same as a normal IoC container.

So what’s different?

Rather than manage these dependencies via an IoC container. I think you should push them in manually. Yes, I’m suggesting you write your own dead simple dependency injection framework. What? Am I mad? Quite probably, but bear with me.

public interface ICanHazTradeDatabase {
    void setTradeDatabase(ITradeDatabase tradeDatabase);
}

public class TradeEntryController
    implements ICanHazTradeDatabase, ICanHazAssetDatabase...
{
    ...
}

public class ControllerFactory {
    public Controller createController(Class clazz) {
        Controller c = clazz.newInstance();
        if (c instanceof ICanHazTradeDatabase)
            ((ICanHazTradeDatabase) c).setTradeDatabase(tradeDatabase);
        if (c instanceof ICanHazAssetDatabase)
            ((ICanHazAssetDatabase) c).setAssetDatabase(assetDatabase);
        if ...

        return c;
    }
}

The exact mechanics of ControllerFactory of course depend on your MVC framework, but hopefully the idea is clear: when we instantiate a controller, we check it against a known set of interfaces and push in very specific dependencies. Is it pretty? Not really. Is it easy to write? Of course. Does it push dependencies into your controller? Well, yes. Where do they come from? Well, that’s an exercise for the reader. But I’m sure you can find a way to make ControllerFactory a singleton and instantiate all your dependencies in one place.

The Point

What, exactly, is the point of all this? Well, as a developer writing a controller – I can get easy access to all the dependencies that represent the architectural context I’m running within. The databases, services, message brokers, email server, blah blah blah that the application as a whole depends on. They’re right there – I just add the interface, one method and bang – ICanHazCheeseburger.

More interesting, is what I can’t do. I can’t decide that my TradeEntryController needs a TradePricingCalculator and inject that as a dependency. Well, I could, but I’d be making TradePricingCalculator available everywhere, and I’ve got a little more work to do than I would if I was using plain old Spring or Guice – I’ve an interface to create, a couple of lines to add to some scarily named GlobalControllerFactory. Why is this important? It adds some friction. It makes something bad hard to do. I’m forced instead to think about creating a TradePrices object and adding some functionality to it. I’m forced to have a rich domain, because I can’t just move all my functionality off into a TradePriceCalculatorVisitorFactoryManagerBuilder.

The choices we make and the technologies we choose make some things easy and other things hard. We need to think carefully about whether the things we make easy should be easy. It’s always possible to do the right thing, but sometimes we need to make it easier than doing the wrong thing.