Home > Articles > Data

A Crash Course in Test-Driven Development

  • Print
  • + Share This
Get a crash course in test-driven development, including the test-first technique, a test's proper role in your organization, what you want to specify, and traditional database design concepts.

The companion code for this chapter can be found here.

This chapter is from the book

Read Test-Driven Database Development: Unlocking Agility or more than 24,000 other books and videos on Safari Books Online. Start a free trial today.



This chapter gives you a crash course in test-driven development (TDD) in case you are not familiar with the discipline.

A staple of the TDD process is the test-first technique. Many people who are new to test-driven development actually confuse it with the test-first technique, but they are not the same thing. Test-first is one tool in the TDD toolbelt, and a very important one at that, but there is a lot more to TDD.

The chapter then covers a test’s proper role in your organization. Tests are best thought of as executable specifications. That is, they not only test something but they also document what that thing should do or how it should look.

One very powerful benefit of cyclically defining and satisfying executable specifications is that it forces your design to emerge incrementally. Each new test you write demands that you revisit and, if necessary, revise your design.

Following that discussion, I cover what you actually want to specify and, probably at least as important, what you do not want to specify. In a nutshell, the rule is “specify behaviors only.” Deciding what a database’s behavior should be can be a little difficult, and I cover that topic in Chapters 6, “Defining Behaviors,” and 7, “Building for Maintainability.” This chapter deals with the behaviors inherent in tables.

Finally, an important piece of test-driven development is to drive behaviors into a database from outside, not the other way around. Again, you can find a lot more advice on how a database should actually be structured later in the book. This chapter deals only with traditional design concepts.

The Test-First Technique

If I were in an elevator, traveling to the top of a building with a software developer I would never see again who had never heard of TDD or the test-first technique, I would try to teach him the test-first technique. I would choose that because it is so easy to teach and it is so easy to get people to try. Also, if done blindly, it creates problems that will force someone to teach himself test-driven development.

The technique is simple, and the following is often enough to teach it:

  1. Write a test.
  2. See it fail.
  3. Make it pass.
  4. Repeat.

There’s nothing more to test-first. There’s a lot more to test-driven development, but test-first really is that simple.

Write the Test

The first step in the technique is to write your test. If you’ve never done this before it might be a little bit uncomfortable at first. You might be thinking “How do I know what to test if there’s nothing there?” That’s a pretty normal feeling that I want you to ball up really tightly and shove down into your gut while you do this a few times. Later you will discover that the best way to determine what should be tested is to write the test for it, but convincing you of that is hard; you’ll have to convince yourself by way of experience.

Anyway, start out by writing a test. Let’s say that I want a database that can store messages sent between users identified by email addresses. The first thing I would do is write a test that requires that ability to be there in order to pass. The test is going to need to create a database of the current version, connect to it, and insert a record. This test is shown in the following listing as one would write it using NUnit and .NET:

[TestFixture]
public class TestFirst {
  private Instantiator instantiator;
  private IDbConnection connection;

  [SetUp]
  public void EstablishConnectionAndRecycleDatabase() {
    instantiator = Instantiator.GetInstance(
      DatabaseDescriptor.LoadFromFile("TestFirstDatabase.xml"));
    connection = DatabaseProvisioning.CreateFreshDatabaseAndConnect();
  }

  [TearDown]
  public void CloseConnection() {
    connection.Close();
  }

  [Test]
  public void TestTables() {
    instantiator.UpgradeToLatestVersion(connection);
    connection.ExecuteSql("
      INSERT INTO USERS VALUES(1, 'foo@bar.com')");
    connection.ExecuteSql(
      @"INSERT INTO MESSAGES " +
      "VALUES(1, 'Hey!', 'Just checking in to see how it''s going.')");
  }
}

That code, as is, won’t compile because I delegate to a little bit of infrastructure that has to be written. One such tool is the DatabaseProvisioning class, which is responsible for creating, tearing down, and connecting to test databases. This class is shown in the following example code, assuming I wanted to test against a SQL Server database:

 public class DatabaseProvisioning {
  public static IDbConnection CreateFreshDatabaseAndConnect() {
    var connection = new SqlConnection(@"Data Source=.\sqlexpress;" +
      "Initial Catalog=master;Integrated Security=True");
    connection.Open();
    connection.ExecuteSql("ALTER DATABASE TDDD_Examples SET " +
      "SINGLE_USER WITH ROLLBACK IMMEDIATE");
    connection.ExecuteSql("DROP DATABASE TDDD_Examples");
    connection.ExecuteSql("CREATE DATABASE TDDD_Examples");
    connection.ExecuteSql("USE TDDD_Examples");

    return connection;
  }
}

The other piece of infrastructure (following) is a small extension class that makes executing SQL statements—something I’m going to be doing a lot in this book—a little easier. For those of you who aren’t C# programmers, what this does is make it look like there is an ExecuteSql method for all instances of IDbConnection.

public static class CommandUtilities {
  public static void ExecuteSql(
    this IDbConnection connection, string toExecute) {
    using (var command = connection.CreateCommand()) {
      command.CommandText = toExecute;
      command.ExecuteNonQuery();
    }
  }
}

The next step is to see a failure.

Stub Out Enough to See a Failure

I like my failures to be interesting. It’s not strictly required, but there’s not a really good reason to avoid it, so assume that making a failure meaningful is implied in “see the test fail.” The main reason you want to see a test fail is because you want to know that it isn’t giving you a false positive. A test that can’t fail for a good reason is about as useful as a test that cannot fail for any reason.

The test I have would fail because there is no database to make, which isn’t a very interesting reason to fail. So let’s create a database class and make it so that the database gets created.

<Database>
  <Version Number="1">
  </Version>
</Database>

With that change in place, my test would fail for an interesting reason: The table into which I was trying to insert doesn’t exist. That’s a meaningful enough failure for me.

See the Test Pass

Now that a test is giving me a worthwhile failure, it’s time to make it pass. I do that by changing the class of databases to create the required table. If I had committed the most recent version of the database class to production, I would create a new version to preserve the integrity of my database class. As it stands, because this new database class hasn’t ever been deployed in an irreversible way, I’ll just update the most recent version to do what I want it to do.

<Database>
  <Version Number="1">
    <Script>
      <![CDATA[
CREATE TABLE Users(ID INT PRIMARY KEY, Email NVARCHAR(4000));

CREATE TABLE Messages(
  UserID INT FOREIGN KEY REFERENCES Users(ID),
  Title NVARCHAR(256),
  Body TEXT);
]]>
    </Script>
  </Version>
</Database>

That update causes my database class to create the message table in version 1. When I rerun my test, the database gets rebuilt with the appropriate structures required to make the test pass. Now I’m done with a test-first programming cycle.

Repeat

After the cycle is complete, there is an opportunity to start another cycle or to do some other things, such as refactoring. I’m going to go through one cycle just to show you how a design can emerge incrementally. After thinking about the design I created, I decided I don’t like it. I don’t want the email addresses to be duplicated.

How should I handle that? I’ll start by adding a test.

[Test]
public void UsersCannotBeDuplicated() {
  instantiator.UpgradeToLatestVersion(connection);
  connection.ExecuteSql(
    @"INSERT INTO Users(Email) VALUES('foo@bar.com')");
  try {
    connection.ExecuteSql(
      @"INSERT INTO Users(Email) VALUES('foo@bar.com')");
  } catch {
    return;
  }

  Assert.Fail("Multiple copies of same email were allowed");
}

After I get that compiling, I’ll watch it fail. It will fail because I can have as many records with a well-known email address as I want. That’s an interesting failure, so I can go on to the next step: adding the constraint to the new version of my database.

<Database>
  <Version Number="1">
    <Script>
      <![CDATA[
CREATE TABLE Users(ID INT PRIMARY KEY, Email NVARCHAR(4000));

ALTER TABLE Users ADD CONSTRAINT OnlyOneEmail UNIQUE (Email);

CREATE TABLE Messages(
  UserID INT FOREIGN KEY REFERENCES Users(ID),
  Title NVARCHAR(256),
  Body TEXT);
]]>
    </Script>
  </Version>
</Database>

Recompiling and rerunning my test shows me that it passes. Had that new behavior caused another test to fail, I would update that test to work with the new design constraint, rerun my tests, and see everything pass. After I’ve done that, I decide I’m done with this phase of updating my database class’s design and move on to other activities.

  • + Share This
  • 🔖 Save To Your Account