Are BDD tests acceptance tests?

问题:

Or, if you have BDD tests, do you need something like Fitnesse?

回答1:

BDD "tests" exist at multiple different levels of granularity, all the way up to the initial project vision. Most people know about the scenarios. A few people remember that BDD started off with the word "should" as a replacement for JUnit's "test" - as a replacement for TDD. The reason I put "tests" in quotes is because BDD isn't really about testing; it's focused on finding the places where there's a lack or mismatch of understanding.

Because of that focus, the conversations are much more important than the BDD tools.

I'm going to say that again. The conversations are much more important than the BDD tools.

Acceptance testing doesn't actually mandate the conversations, and usually works from the assumption that the tests you're writing are the right tests. In BDD we assume that we don't know what we're doing (and probably don't know that we don't know). This is why we use things like "Given, When, Then" - so that we can have conversations around the scenarios and / or unit-level examples. (Those are the two levels most people are familiar with - the equivalent of acceptance tests and unit tests - but it goes up the scale).

We don't call them "acceptance tests" because you can't ask a business person "Please help me with my acceptance test". They'll look at you with a really weird, squinty gaze and then dismiss you as that geek girl. 93% of you don't want that.

Try "I'd like to talk to you about the scenario where..." instead. Or, "Can you give me an example?" Either of these are good. Calling them "Acceptance tests" starts making people think that you're actually doing testing, which would imply that you know what you're doing and just want to make sure you've done it. At that point, conversations tend to focus on how quickly you can get the wrong thing out, instead of about the fact you're getting out the wrong thing.

And you're getting the wrong thing out. Really, honestly, you are. Even if you think you're not, it's only because you don't understand second-order ignorance. You don't know that you don't know, and that's OK, as long as you've found the places where you could know you don't know. (You won't find all of them. Don't let the categorisation paradox keep you up at night.)

The only way to really get it right is to get all the requirements up front, and you know what happens when you try that. That's right. It's Waterfall. Remember the overtime? The weekend work? The seven years in which not one thing you created made it to production? If you want to avoid that, you only have one chance: assume you're wrong, have some conversations about it to be less wrong, then accept that you're still wrong and go for it anyway. Writing tests too early means you have even more chance to be wrong, and now it's harder to change and everyone thinks you're right and the PM is measuring your velocity and now you're committed to being wrong for another 2 weeks. And - worse - you're about to test that you're wrong, too.

Once again. The conversations are much more important than the BDD tools.

Please, please, don't fixate on the tools. The tools are just a mechanism for capturing the conversations and making sure that they get played into the code. Scenarios are not a replacement for conversations, any more than a 3 x 5 index card is a replacement for requirements.

Having said that, if you must start with a tool, put Slim behind Fitnesse so that it can run lovely Given / When / Thens without having to mess with Fit's tables and fixtures. GivWenZen is based on Slim and either of them rocks. FitSharp is the equivalent for those of you in the .NET space. Or just use Cucumber, or SpecFlow, or knock up a little custom DSL* that will do the job fine for years.

Transparency: *I wrote that one. And bits of JBehave. I wish we had called it "Dont-concentrate-on-BDD-tools-Behave". I might be heavily involved in other bits of BDD. Plus Dan North will buy me a pint if I can get this message out, so it's not exactly impartial advice.

Regardless - have the conversations already. It's just people. Go talk.

回答2:

I don't know if there's such a thing, strictly speaking, as a "BDD test". BDD is a philosophy that suggests how you can best interact and collaborate with stakeholders to complete a complex project. It doesn't directly make any prescriptions for the best way to write tests. In other words, you'll probably still have all the usual kinds of tests (including acceptance tests) under a BDD-philosophy project.

When you hear of "BDD frameworks", the speaker usually means a framework for writing all your usual kinds of tests but with a BDD twist. For example, in RSpec, you still write unit tests; you just add the BDD flavor to them.

回答3:

I like to draw a distinction between "specs" and "tests."

If I am covering a method called GetAverage(IEnumerable<int> numbers), I am going to write a more or less standard unit test.

If I am covering a method called CalculateSalesTax(decimal amount, State state, Municipality municipality), I am still going to write the unit test, but I'll call it a specification because I'm going to change it up (1) to verify the behaviour of the routine, and (2) because the test itself will document both the routine and its acceptance criteria.

Consider:

[TestFixture]
public class When_told_to_calculate_sales_tax_for_a_given_state_and_municipality() // the name defines the context
{
    // set up mocks and expected behaviour
    StateSalesTaxWebService stateService 
        = the_dependency<IStateSalesTaxWebService>;
    MunicipalSurchargesWebService municipalService 
        = the_dependency<IMunicipalSurchargesWebService>;

   stateService.Stub(x => x.GetTaxRate(State.Florida))
        .Return(0.6);
    municipalService.Stub(x => x.GetSurchargeRate(Municipality.OrangeCounty))
        .Return(0.05);

    // run what's being tested
    decimal result = new SalesTaxCalculator().CalculateSalesTax
        (10m, State.Florida, Municipality.OrangeCounty);

    // verify the expected behaviour (these are the specifications)
    [Test]
    public void should_check_the_state_sales_tax_rate()
    {
        stateService.was_told_to(x => x.GetTaxRate(State.Florida)); // extension methods wrap assertions
    }

    [Test]
    public void should_check_the_municipal_surcharge_rate()
    {
        municipalService.was_told_to(x => x.GetSurchargeRate(Municipality.OrangeCounty));
    }

    [Test]
    public void should_return_the_correct_sales_tax_amount()
    {
        result.should_be_equal_to(10.65m);
    }
}

回答4:

JBehave (and NBehave recently added the same support) work with regular test files so while many other frameworks add "BDD taste tounit tests" the text based behaviour specifications/examples created with JBehave are suitable for acceptance tests. And no, you don't need fitnesse for that.

To get an idea of how it works I suggest JBehaves 2min tutorial.

回答5:

While BDD is larger than the scope of just tests, there are indeed BDD tests. These tests are Unit Tests that follow the BDD language.

Given some initial context (the givens), When an event occurs, then ensure some outcomes.

There are a few good BDD frameworks available depending on your language of preference. JBehave for Java RSpec for Ruby NBehave for .NET

回答6:

For BDD testing in Flex you can try GivWenZen-flex check it out http://bitbucket.org/loomis/givwenzen-flex.

Cheers, Kris

回答7:

xBehavior BDD tests implemented well are robo-driven user acceptance criteria.

xSpecification BDD tests are normally unit tests and are unlikely to be acceptable user acceptance criteria.