I admit, up until now I’ve largely ignored the BDD (Behavior Driven Development) Context/Specification style of writing unit tests. It’s been touted as a more approachable way to learn TDD (Test Driven Development) and as a more natural transition from user stories to the actual code design. I guess my hesitation to give it a second thought was that I felt I didn’t need a more approachable form of TDD.

Recently, my Subtext partner in crime, Steve Harman, urged me to take another fresh look at BDD Context/Specification style tests. I trust Steve’s opinion, so I took another look and in doing so, the benefits of this approach dawned on me. I realized that it wasn’t BDD itself I didn’t like, after all, I did enjoy writing specs using minispec and IronRuby. I realized that the part I didn’t really like was the .NET implementations of this style. Keep in mind that I do not claim to be an expert in TDD or BDD, I’m just a student and these are just my observations. I’m sure others will chime in and provide corrections that we can all learn from.

SpecUnit.NET example

For example, let’s take a look at one example pulled from the sample project of Scott Bellware’s SpecUnit.NET project, which provides extensions supporting the BDD-style use with .NET unit testing frameworks and has really pushed this space forward. I trimmed the name of the class slightly by removing a couple articles (“the” and “an”) so it would fit within the format of my blog post.

[Concern("Funds transfer")]
public class when_transfering_amount_greater_than_balance_of_the_from_account
  : behaves_like_context_with_from_account_and_to_account
{
  private Exception _exception;

  protected override void Because()
  {
    _exception = ((MethodThatThrows) delegate
    {
      _fromAccount.Transfer(2m, _toAccount);
    })
    .GetException();
  }

  [Observation]
  public void should_not_allow_the_transfer()
  {
    _exception.ShouldNotBeNull();
  }

  [Observation]
  public void should_raise_System_Exception()
  {
    _exception.ShouldBeOfType(typeof(Exception));
  }
}

The Because method contains the code with the behavior we’re interested in testing. The two methods annotated with Observation are the specifications. Notice that the names of the classes and methods are meant to be human readable. The output of running these tests remove underscores and reads like a specification document. It’s all very cool.

What I like about this approach is there’s a crisp focus on having each test class focused on a single behavior, in this case transferring a balance from one account to another. In the past, I might have written something like this as two test methods (which led to duplicating code or putting code in some generic Setup method that seems detached from what I’m trying to test) or as one method with two asserts. This approach helps you think about how to organize tests along the lines of your objects’ responsibilities.

What I don’t like about it, and I admit this is really just a nitpick, is that it looks like someone’s keyboard puked underscores all over the place. I feel like having to encapsulate each observation as a method adds a lot of syntactic overhead when I’m trying to read this class from top to bottom. Maybe that’s just something you get used to.

MSpec example

Switching gears, let’s look at a different example by Aaron Jensen. This is an experiment in which he tried a very different approach. Look at this code sample…

[Description]   
public class Transferring_between_from_account_and_to_account   
{   
  static Account fromAccount;   
  static Account toAccount;   
  
  Context before_each =()=>   
  {   
    fromAccount = new Account {Balance = 1m};   
    toAccount = new Account {Balance = 1m};   
  };   
     
  When the_transfer_is_made =()=>   
  {   
    fromAccount.Transfer(1m, toAccount);   
  };   
      
  It should_debit_the_from_account_by_the_amount_transferred =()=>   
  {   
    fromAccount.Balance.ShouldEqual(0m);   
  };   
  
  It should_credit_the_to_account_by_the_amount_transferred =()=>   
  {   
    toAccount.Balance.ShouldEqual(2m);   
  };   
}  

There’s still the underscore porn, but it does read a little more like prose from top to bottom, if you can get yourself to ignore that funky operator right there. =()=> Whoa!

When I complained to Steve about all the underscores in these various approaches, he suggested that being a fan of the more Zen-like Ruby language, the underscores didn’t bother him. I didn’t buy that as part of the aesthetic of Ruby is its clean DRY minimalism. Yes, it uses underscores, but it doesn’t generally abuse them. Let’s take a look at a BDD example using RSpec and Ruby. This is an example of a spec in progress from Luke Redpath… (forgive the poor syntax highlighting. I need a ruby css stylesheet. :)

context "A user (in general)" do 
  setup do 
    @user = User.new 
  end 

  specify "should be invalid without a username" do 
    @user.should_not_be_valid 
    @user.errors.on(:username).should_equal "is required" 
    @user.username = 'someusername' 
    @user.should_be_valid 
  end 

  specify "should be invalid without an email" do 
    @user.should_not_be_valid 
    @user.errors.on(:email).should_equal "is required" 
    @user.email = 'joe@bloggs.com' 
    @user.should_be_valid 
  end 
end

One thing to notice is that we’re not using separate methods and classes here. Ruby doesn’t force you to put code in classes. You can just execute a script top-to-bottom. In this case, the code sets up a context block and within that block there is a setup block and a couple of specify blocks. There’s no need to factor a specification into multiple classes and methods.

Also notice that the context and specifications are described using strings! Now we’re getting somewhere. If it’s meant to be human readable, why don’t we use strings instead of the underscore porn? On Twitter, many accused the ceremony and vagaries of C# of preventing this approach. While I agree that Ruby has less ceremony than C#, I also think C# doesn’t get its fair shake sometimes. We can certainly take a C# approach down to its bare metal with the least syntactic noise as possible, right?

SubSpec

So in true Program Manager at Microsoft fashion, I spec’d out a rough idea of the syntax I would like to use with BDD. I then showed it to Brad Wilson asking him how can I make this work in xUnit.net. In true Developer fashion, he ran with it and made it actually work. This blog post is the part where I try to take all the credit. That’s what PMs do at Microsoft, write specs, take credit for the hard work the developers do in bringing the specs to life. ;) (I kid, I kid)

Here’s an example using this syntax…

[Specification]
public void PushNullSpecifications()
{
  Stack<string> stack = null;

  "Given a new stack".Context(() => stack = new Stack<string>());

  "with null pushed into it".Do(() => stack.Push(null));

  "the stack is not empty".Assert(() => Assert.False(stack.IsEmpty));
  "the popped value is null".Assert(() => Assert.Null(stack.Pop()));
  "Top returns null".Assert(() => Assert.Null(stack.Top));
}

A few things to notice. First, the entire spec is in a single method, which reduces some of the syntactic noise of splitting the spec into multiple methods. Secondly, we’re using strings here to describe the specification and context, rather than method names with underscores for the human readable part.

Lastly, and most importantly, while it may look like we’re committing the sin of multiple asserts in a single test, this is not the case. Via the power of the xUnit.NET extensibility model, Brad was able to generate a test per assertion. That’s why the Assert method (should it be Observe or Fact?) takes in a lambda. We can return these closures to xUnit.net and it will wrap each one in a separate test. Here’s another look at the same method with some comments to highlight how similar this is to the previous examples. (UPDATE: I also changed the asserts to use the Should style extension methods to demonstrate what it could look like).

[Specification]
public void PushNullSpecifications()
{
  Stack<string> stack = null;
  //equivalent to before-each
  "Given a new stack".Context(() => stack = new Stack<string>());

  //equivalent to Because()
  "with null pushed into it".Do(() => stack.Push(null));

  //Equivalent to [Observation]
  "the stack is not empty".Assert(() => stack.IsEmpty.ShouldBeFalse());
  //Equivalent to [Observation]
  "the popped value is null".Assert(() => stack.Pop().ShouldBeNull());
  //Equivalent to [Observation]
  "Top returns null".Assert(() => stack.Top.ShouldBeNull());
}

Keep in mind, at this point, this is a merely proof-of-concept sample that will be included with the samples project in the next version of xUnit.NET and by the time you read this sentence, it may have changed already. You can download this particular changeset here.

The following is a screenshot of the HTML report generated by xUnit.NET when using this syntax that Brad sent me today.subspec-report

Despite it being a sample, I tried to give it a catchy name in case this is intriguing to others and worth iterating on to make it better (not to mention that I love putting the prefix “Sub” in front of everything.)

Possible next steps would be to add all the Woulda, Coulda, Shoulda extension methods so popular with this style of testing. For example, that would allow you to replace Assert.False(stack.IsEmpty) with stack.IsEmpty.ShouldBeFalse(). For those of you practicing BDD, I’d be interested in hearing your thoughts, objections, etc… concerning this approach.

For completeness sake, here’s another syntax I proposed to Brad. He mentioned it was similar to something else he’s seen which he might port over to xUnit.net.

[Specification]
public void When_Transferring_To_An_Account()
{
  Election e = null;
  Account a = null;
  Account b = null;
 
  Where("both accounts have positive balances", () => {
    a = new Account { Balance = 1 };
    b = new Account { Balance = 2 };
  });
 
  When("transfer is made", () =>
    
    a.Transfer(1, b)
  );
 
  It("debits account by amount transferred", () => a.Balance.ShouldBe(0));
  It("credits account by amount transferred", () => b.Balance.ShouldBe(3));
}

For those of you completely new to BDD, check out Scott Bellware’s Code Magazine article on the subject.