Unit testing Sitecore MVC

800px-Nile_blue_05

ASP.NET MVC and unit testing go hand in hand. The basic MVC architecture forces a separation of concerns that lends itself well to unit testing, and likewise so for the most commonly used patterns (like repositories, unit of work, and business logic ‘domains’).

A common scenario in standard ASP.NET MVC

I think that anyone wanting to learn Sitecore MVC should take some time to understand pure ASP.NET MVC as well. It’s important to understand where Sitecore does things differently – and why, and what effect that has.

A common scenario in standard ASP.NET MVC might look like this. A request comes in for /Sample/Index/30. This request is routed to a controller first of all:

  1. A controller should be light, and is responsible for assembling your view model (or just models, to keep things simple). Models are dumb classes with no business or data-fetching logic. This layer asks your business logic layer for an object.
  2. A business logic layer (sometimes called a ‘domain’) is responsible for all your business logic. This retrieves the object from the repository, and does any additional business logic.
  3. A repository is responsible for fetching your data – there should be no business logic. This layer only returns data from whatever your data source is.
  4. The controller, having put together the model or view model, returns the view. Views are responsible for display, and should contain minimal logic.

Often, dependency injection will be used at the root of the application to ensure that no layers are dependent on one another. For example, the controller will not new up a concrete implementation of IBusinessLogic; it will be injected when the controller is instantiated.

Let’s see what this looks like in practice. I’ve got a controller that is responsible for ‘Sample’ objects. /Sample/Index/30 maps to the Index action result, and 30 is passed into the id parameter:

namespace Simple_MVC.Controllers
{
    public class SampleController : Controller
    {
        // Inject the dependencies - this controller needs access to the business logic layer
        private readonly IBusinessLogic _businessLogic;
        public SampleController(IBusinessLogic businessLogic)
        {
            _businessLogic = businessLogic;
        }

        public ActionResult Index(int id)
        {
            // Go and fetch the sample object with the specified ID
            var sample = _businessLogic.GetSample(id);

            if (sample == null)
            {
                return RedirectToAction("Error");
            }

            return View(sample);
        }

        public ActionResult Error()
        {
            return View();
        }
    }
}
  • The concrete implementation for IBusinessLogic is injected by Castle Windsor when the controller is instantiated – but you can use any Inversion of Control container
  • The controller is light – it is only responsible for putting together my model (with should really be a view model)
  • The business logic happens in the concrete IBusinessLogic – if I needed to swap my business logic layer, I could just inject something else

The business logic layer I am using is very simple – it checks that the ID coming into the Index action is greater than 20:

 public class StrictBusinessLogic : IBusinessLogic
    {
        // Inject the dependencies - this controller needs access to the data-fetching repository layer
        IRepository _repository;
        public StrictBusinessLogic(IRepository repository)
        {
            _repository = repository;
        }
        // Business logic check - ID must be greater than 20. If yes, go and fetch the data.
        public Sample GetSample(int id)
        {
            if (id >= 20)
            {
                return _repository.GetSampleObject(id);
            }

            return null;
        }
    }
  • This layer has a dependency on a data-fetching repository layer, which is injected by Castle Windsor.
  • There is only business logic in this layer – it has no direct access to the data source (which is usually a database).
public class SampleRepository : IRepository
    {
        // Sample repository layer - no business logic
        public Models.Sample GetSampleObject(int id)
        {
            Sample sample = new Sample();

            sample.ID = id;
            sample.Title = "Sample Repository Title";
            sample.Text = "Sample Repository Text";

            return sample;
        }
    }
  • The repository returns data – it contains no business logic.
  • Usually, ASP.NET MVC is used together with an ORM such as Entity Framework – the GetSampleObject would call an ORM-specific method that retrieves all Sample objects that match the specified ID.

Seems like a lot of layers – what am I gaining?

First and foremost, the separation of concerns enforced by MVC makes it very testable – but there are other gains to be had from the modularity of the MVC approach.

Testability

I can test every part of this implementation individually – from the controller to the business logic layer. If I want to test the controller behaviour, I can either mock the IBusinessLogic dependency or create an instance of it as a concrete implementation (which would in turn require a fake implementation of IRepository).

In the unit test example below, the Moq library has been used to fake the IBusinessLogic layer:

[TestMethod]
public void ControllerRedirectsToError()
{
    var businessLogicFake = new Mock<IBusinessLogic>();

    businessLogicFake.Setup(x => x.GetSample(3)).Returns((Sample)null);
         
    var controller = new SampleController(businessLogicFake.Object);

    var result = controller.Index(3) as RedirectToRouteResult;

    Assert.IsTrue(result.RouteValues.ContainsKey("action"));

    Assert.AreEqual("Error", result.RouteValues["action"].ToString());
}

Or, if you do not want to use a mocking library:

[TestMethod]
public void ControllerRedirectsToError_WithoutMocking()
{
    // A fake data access layer - no depenency on a database
    var repository = new SampleRepository();

    var businessLogicFake = new StrictBusinessLogic(repository);

    var controller = new SampleController(businessLogicFake);

    var result = controller.Index(3) as RedirectToRouteResult;

    Assert.IsTrue(result.RouteValues.ContainsKey("action"));

    Assert.AreEqual("Error", result.RouteValues["action"].ToString());
}

Note! This post is not about unit testing best practices; be gentle when assessing the ‘usefulness’ of these tests, or how well they are written. 🙂

But there’s more!

If a particular section is under construction or with another team, I can inject a fake, temporary piece to fill that dependency whilst they are working. When that piece is complete, I inject the completed code.

In addition, I know that each layer has a very particular responsibility. As a developer, I know exactly where I have to go for a particular problem – is something going wrong with my special offers calculation? Check business logic layer first. Is there a problem with the appearance of the page? Check the view.

Back to Sitecore

Sitecore will not stop you from structuring your application in the manner detailed above if you are using controller renderings. In fact, that entire code sample could almost be a controller rendering – you can use dependency injection and you can move your business logic into a separate layer.

On the controller rendering vs view rendering debate, I can only say this – if I need to do something complicated that requires business logic of any kind, I use a controller rendering. If I need to render some field data, I use a view rendering. Views should contain minimal logic, so I am quite happy to leave those untested.

OK – then what exactly is stopping me from unit testing Sitecore?

Have a look at this very simple controller:

public class CircuitController : Controller
    {
        RenderingContext _renderingContext;
        public CircuitController(RenderingContext renderingContext)
        { 
             _renderingContext = renderingContext;        
        }

        public CircuitController() : this(RenderingContext.Current)
        {
        }

        public ActionResult Featured()
        {            
            Circuit circuit = new Circuit();

            circuit.Initialize(_renderingContext);

            return View(circuit);
        }
    }
}

The controller above takes a RenderingContext object. Unfortunately RenderingContext does not have an interface, so you cannot inject the concrete implementation. If a a RenderingContext is not provided, the empty constructor will fire with a default – RenderingContext.Current, which relies on Sitecore. This isn’t true dependency injection (dependency injection should be done in the root of your application by your IoC container), but it allows us to test.

So far so good – the empty constructor will work in a Sitecore context (when RenderingContext.Current is available), and I can pass in my own RenderingContext object in testing.

Let’s try writing a test. The test won’t actually do anything – but it should still run:

        [TestMethod]
        public void CircuitControllerTest_WillNeverWork()
        {
            CircuitController circuitController = new CircuitController( … );

            var result = circuitController.Featured();
        }

The CircuitController needs a RenderingContext; if one isn’t provided, it will try to access Sitecore – which we cannot do in the context of a test. Can I mock one? At first glance, yes, but the Current property required by my controller is static, and it only has a getter.

Can I create my own version from an interface? No, we’ve just said it doesn’t have one.

So… what do I do? Exactly.

You will come across issues like this every time you touch the Sitecore API – the Item is dependent on receiving a Database, and that in turn has a dependency on a configuration file. There are static classes everywhere, and precious few interfaces.

But before we lose ourselves in complaining, it is worth remembering that Sitecore is no longer a fresh, greenfield baby – the company is thirteen years old this year, and so is the codebase. The newer parts of the API (notably search) are full of interfaces, and are very testable.

There are also ways to get around the problems we’ve just encountered.

Wrap and isolate

Unit testing Sitecore is an exercise in isolation.

If Sitecore makes our application difficult to test, move any references to the API out of your controllers and business logic, and wrap any classes (like Item) that you do need to use. Let’s start with wrapping.

Wrapping Sitecore classes

Have a look at this example:

    public class LocationController : Controller
    {
        private readonly ILocationDomain _locationDomain;
        private readonly IPageContext _pageContext;
        private readonly IRenderingContext _renderingContext;
        public LocationController(ILocationDomain locationDomain, IPageContext pageContext, IRenderingContext renderingContext)
        {
            _locationDomain = locationDomain;
            _pageContext = pageContext;
            _renderingContext = renderingContext;
        }

        public ActionResult Featured()
        {
            LocationViewModel viewModel = new LocationViewModel();

            viewModel.Location = _locationDomain.GetLocation(_renderingContext.Rendering.Item);

            if (viewModel.Location == null)
            {
                if (_pageContext.IsPageEditor)
                {
                    return View("~/Views/Shared/_NoDatasource.cshtml");
                }

                return new EmptyResult();
            }

            return View(viewModel);
        }
    }

This is a controller rendering for a featured location. We are injecting our business logic (ILocationDomain), but we are also injecting something called an IPageContext and an IRenderingContext. Are those Sitecore interfaces?

No – those are interfaces I have created myself. Let’s take the IPageContext as an example:

public interface IPageContext
    {
        IItem Current { get; }
        bool IsPageEditor { get; }
    }

And here is a concrete implementation:

public class PageContextWrapper : IPageContext
    {
        public IItem Current
        {
            get
            {
                return new ItemWrapper(PageContext.CurrentOrNull.Item);
            }
        }

        public bool IsPageEditor
        {
            get
            {
                return Sitecore.Context.PageMode.IsPageEditor;
            }
        }
    }

You’ll notice that an IItem interface is referenced – that is also one of mine:

public interface IItem
    {
        string DisplayName { get; }
        ID TemplateID { get; }
        Item Item { get; set; }
    }

The concrete implementation looks like this:

public class ItemWrapper : IItem
    {
        public ItemWrapper(Item item)
        {
            Item = item;
        }

        public ItemWrapper()
        {
        }

        public Item Item { get; set; }

        public string DisplayName
        {
            get
            {
                return Item.DisplayName;
            }
        }

        public ID TemplateID
        {
            get
            {
                return Item.TemplateID;
            }
        }
    }

My code will never call the Sitecore API directly. It will always go via these wrappers. Take this line from the controller as an example:

 if (_pageContext.IsPageEditor)

In a Sitecore context, that will hit Sitecore.Context.PageMode.IsPageEditor and return true or false depending on whether or not we are in the Page Editor. However, if I wanted to test my controller, I could create a fake implementation of IPageContext – or mock that interface. The unit testing code will never hit the Sitecore API, and therefore does not have a dependency on it.

My IRenderingContext is passed on to the business layer, which similarly does not touch the Sitecore API – in addition, any calls that layers needs to make to methods such as FieldRenderer.Render() are safely tucked away in a Sitecore-specific repository, and injected.

You will find yourself having to do quite a lot of wrapping. In the sample project, I have an entire folder of wrappers – Item, PageContext, Rendering, and so on – but it does mean that you can perform a true unit test.

Isolating the Sitecore API

In addition to wrapping Sitecore classes, you can also isolate the Sitecore API away from your code by putting it in a repository. For example – in my business logic layer, I want to do some checks before returning the featured location that my controller is requesting. Remember – there should be no data-fetching in this layer at all (and in this instance, our ‘database’ is Sitecore).

My business logic layer injects a data-fetching repository:

public class LocationDomain : MVC.Data.Domain.ILocationDomain
    {
        private readonly ISitecoreRepository _sitecoreRepository; 
        public LocationDomain(ISitecoreRepository sitecoreRepository)
        {
            _sitecoreRepository = sitecoreRepository;
        }

All and any Sitecore API calls go in this repository – and because it is injected, I can inject a fake one during testing. Let’s have a look at where the repository is used within the business logic layer. The method below is responsible for returning a Location:

public Location GetLocation(IItem item, Dictionary<string, string> parametersDictionary)
        {
            Location location = null;
            
            if (item != null)
            {
                    if (item.TemplateID == References.LocationTemplateID)
                    {
                        location = new Location();

                        location.Title = new System.Web.HtmlString(GetField("Name", item, parametersDictionary));
                        location.Text = new System.Web.HtmlString(GetField("Text", item, parametersDictionary));

                        return location;
                    }
                }
            
            return null;
        }

It checks if the incoming item’s TemplateID is correct. If yes, it goes on to create a Location, and populates each property with a GetField() method that takes a field name and an ‘item’ (which is actually a wrapper class).

Inside the GetField() method, we have our first reference to the repository:

public string GetField(string fieldName, IItem item, Dictionary<string, string> parameters)
        {
            if (item != null)
            {
                if (!String.IsNullOrEmpty(fieldName))
                {
                    if (_sitecoreRepository.FieldExists(fieldName, item))
                    {
                        string fieldParameters = GetFieldParameters(fieldName, parameters);

                        if (!String.IsNullOrEmpty(fieldParameters))
                        {
                            return _sitecoreRepository.GetFieldValue(fieldName, item, fieldParameters);
                        }
                        else
                        {
                            return _sitecoreRepository.GetFieldValue(fieldName, item);
                        }
                    }
                }         
            }

            return String.Empty;
        }

Not once do we touch the Sitecore API. Every call – including the check to see if a field exists – is done using the repository. The Sitecore implementation of GetFieldValue contains minimal code and no business logic:

public class SitecoreRepository : ISitecoreRepository
    {
        public string GetFieldValue(string fieldName, IItem item, string parameters)
        {
            return FieldRenderer.Render(item.Item, fieldName, parameters);
        }

With Sitecore classes wrapped and Sitecore methods isolated, I no longer have a dependency on Sitecore. In my unit tests, the Sitecore-specific implementations of wrappers can be swapped out for fake ones, as can the Sitecore repository.

Here is a controller unit test (which uses mocking of my own wrapper interfaces). It should return ‘null’ if the business layer does not return a valid Location:

[TestMethod]
public void LocationControllerTest_ReturnsEmptyIfNoLocation()
{
    var repository = Mock.Of<ISitecoreRepository>();
    var pageContext = Mock.Of<IPageContext>();

    var rendering = Mock.Of<IRenderingWrapper>();

    var renderingContext = Mock.Of<IRenderingContext>(rc => rc.Rendering == rendering);

    LocationDomain domain = new LocationDomain(repository);

    LocationController locationController = new LocationController(domain, pageContext, renderingContext);

    var result = locationController.Featured();

    Assert.AreEqual("EmptyResult", result.GetType().Name);
}

Here is a business layer unit test. It should return null if an item does not have the correct template:

[TestMethod()]
        public void GetLocation_NullTemplate()
        {
            var item = Mock.Of<IItem>();

            var repository = Mock.Of<FakeRepository>();

            var domain = new LocationDomain(repository);

            var location = domain.GetLocation(item);

            Assert.IsNull(location);
        }

LocationDomain has a dependency on IRepository, so I am injecting a fake repository instead of the one that contains calls to Sitecore.

Are there any other options?

You do have to take unit testing into account at the very beginning of your Sitecore project. However – you have to take unit testing into account at the beginning of any MVC project, because it directly affects how you structure your code.

There are other ways to unit test Sitecore, including but not limited to:

  • TypeMock, or any other mocking framework that can mock static classes – the free Moq cannot, as far as I’m aware.
  • Use an ORM like Glass on top of Sitecore.
  • Use something like Sitecore Fakes*, which does something similar to what has been described above.
  • Any other ‘wrap and isolate’ methods you can think of

Resources

You can see the complete code sample for the LocationController et al on GitHub: https://github.com/Sitecore-Community/sample-sitecore-mvc

Credits

Top image from: http://commons.wikimedia.org/wiki/File:Nile_blue_05.jpg

One comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s