Debugging .NET with NUnit
Like this article? We recommend
Like this article? We recommend
Like this article? We recommend
What is NUnit? NUnit is a testing concept based on JUnit. Originally a concept for testing that wraps an interface around code, permitting testing tools to call into that code checking for pass and fail conditions. The result is automated testing extraordinaire.
I am working on a project in Oregon that is using NUnit version 2.0 (available from http://www.nunit.org). This open source product has been updated to use the best of .NET for implementing automated and regression tests. You can download, extend, and use NUnit for free.
The product uses custom attributes to guide the tests. In addition, the code being tested is loaded into its own application domain. As a result, the NUnit tool can load and reload the .NET assemblies you are testing without restarting the NUnit tool. The benefit is that NUnit version 2.0 looks for changes to the test assemblies and will reload them automatically to ensure that the latest binary assembly is being tested. As a developer, this means you can perform code and test development cycles, running NUnit with loaded assemblies to be tested after each compilation without restarting and loading test fixtures.
The net result is that NUnit integrates easily into unit testing and .NET in general, and is an excellent tool for system and regression testing. With the simple pass-fail user interface, there is no mistaking tests that failed and tests that passed. Because tests are written in .NET, analysts, programmers, and testers can codify tests and capture them as simple .NET algorithms that can be ascertained to validate the business rules.
In this article, I will introduce the fundamental concepts of .NET and NUnit testing. I also will show you one hint that we—Joe Shook and I—learned the hard way that will help you test even very complex applications.
Implementing a Test
One of the benefits of writing is that you get to state an opinion. The purpose of a stated opinion is sometimes to provide the correct information and other times to provoke a dialog. Here is my provocation: Programmers make the best testers, especially when they act as friendly adversaries that did not write the code being tested. Other professionals have this opinion, too. The reason I restate it here is because the authors of the code were the only testers on more than 75% of the projects I have been on. I have begun to believe that the lack of autonomous testers is both commonplace and an egregious mistake made by managers and budget commandos.
If the originating author tests the code, you will probably get all positive test cases based on assumptions and foreknowledge. These are called unit tests, and if all that is happening is unit testing, some buggy code is getting shipped.
Beneficially, what NUnit does is provide us all with a common tool for testing. Analysts can write both positive and negative test cases, which can be captured and codified in .NET by programmers who did not write the code, and the code either passes or fails. A second huge benefit is that the authors of the code can run the same tests during unit testing, getting a head start on bug-free code. As friendly adversaries, it is the tester's job to keep adding twists. Now your testers get started writing tests about the same time your coders do. Polish all this off with quality assurance—proofreading, formatting, fit and finish—and some great software will get shipped.
To demonstrate how all of this can and will work, I have included a trivial Windows Forms application, a class library representing business rules, and a test assembly that uses NUnit. Let's start with the pseudo-business rules and finish with the test assembly.
Building Sample Business Rules
Test code needs to be in a .DLL assembly, which is representative of the middleware part of your code commonly referred to as business rules. Because our emphasis is on testing, we can use just about any code, so I created a Class Library project containing a class named Arithmetic. Arithmetic performs simple verbose addition for integer and double numeric types (see Listing 1). To create a negative test, I wanted to check overflow errors—when an arithmetic operation causes a number to exceed its maximum value, for example, 3,000,000,000 assigned to an integer contains more than 32 bits. To test an arithmetic overflow, I turned "Check for Arithmetic Overflow/Underflow" on in the Property Pages|Configuration Properties|Build (see Figure 1).
)
Figure 1 Enabling checks for arithmetic underflow and overflow.
Listing 1: Sample business rules we will be testing with NUnit.
using System;
namespace MyMath
{
public class Arithmetic
{
public static int Add(int a, int b)
{
return a + b;
}
public static double Add(double a, double b)
{
return a + b;
}
}
}
The code contains two static addition methods.
TIP
Another good reason for using NUnit is implied. Because we can load class libraries only in NUnit testing, it forces a good separation between the presentation layer (GUI) and business rules. This is a good thing.
The key to testing is defining a test that supports a business rule regardless of the code we are testing; hence we need to ensure an accurate result for addition.
Building a GUI Application
For our purposes, the GUI simply demonstrates how to use the Arithmetic class and represents our presentation layer. The form I created contained three TextBox controls, three labels, and a button. Enter a numeric value in each of the first two TextBoxes, click the button, and the sum is returned in the third TextBox. The MyMath assembly is referenced by the Windows Forms application, and the arithmetic is invoked in the following manner:
textBox3.Text = Arithmetic.Add(Convert.ToInt32(textBox1.Text), Convert.ToInt32(textBox2.Text));
Configuring NUnit
We now have some code to test. We will need to download and install NUnit and then build our test fixture. You can download NUnit from http://www.nunit.org and use all the default values for installation. This will install the source code and binaries in C:\Program Files\NUnit V2.0 by default.
Implementing the Test Assembly
Writing the tests represents the new information for us, so we'll take our time here.
NOTE
Recall that I said that testers should be programmers, but not the authors of the code. Well, it is possible to deviate a little bit when using NUnit. Suitable conditions that permit programmers and testers to be in the same corpus are when an analyst or analysts have defined both positive and negative tests without knowledge of the code. Under these circumstances, the tests aren't directed by the programmer, and you will get good results.
A test is implemented as a Class Library. The test library references your business rules code and invokes operations on that code, just as your presentation layer will be doing.
The way that NUnit works is to load the test .DLL into an AppDomain object, relying on attributes and Reflection to determine which code represents tests. What we must do is apply attributes that NUnit will be looking for and write the tests appropriately. Listing 2 contains some tests I wrote for the MyMath.Arithmetic, followed by an explanation of each of the attributes and the way the tests were written.
Listing 2: Test fixture for MyMath.Airthmetic.
1: using System;
2: using NUnit.Framework;
3: using MyMath;
4:
5: namespace Test
6: {
7: [TestFixture()]
8: public class TestArithmetic
9: {
10:
11: //[SetUp()]
12: public void Init()
13: {
14: // Initialization for each test method here!!!
15: }
16:
17: [TearDown()]
18: public void Deinit()
19: {
20: // De-initialization for each test method here!!
21: }
22:
23: [Test()]
24: public void AddIntegerPass()
25: {
26: Assertion.AssertEquals("Integer arithmetic passed",
27: true, Arithmetic.Add(10, 5) == 15);
28: }
29:
30: [Test()]
31: public void AddDoublePass()
32: { 33: Assertion.AssertEquals("Floating-point arithmetic passed",
34: true, Arithmetic.Add(1.3, 2.5) == 3.8);
35: }
36:
37: [Test(), ExpectedException(typeof(OverflowException))]
38: public void AddIntegerFailed()
39: {
40: int a = 2147483647;
41: int b = 5;
42: Assertion.AssertEquals("Integer overflow expected",
43: true, Arithmetic.Add((int)a, (int)b) > 2147483647);
44: }
45:
46: [Ignore("not ready")]
47: public void AddDoubleFailed()
48: {
49: // not ready!
50: }
51:
52: }
53: }
NUnit attributes and classes that we will be defining are contained in C:\Program Files\NUnit V2.0\bin\NUnit.Framework.dll. We will need to add a reference to NUnit.Framework.dll and MyMath.dll—the code we will be testing. For convenience, I added a using clause to their respective namespaces as well.
To indicate that a class is a test fixture, apply the TestFixtureAttribute to the class, as shown on line 7.
If you need initialization and deinitialization code, add two methods to the test fixture. I named these Init and Deinit, but they can be anything. It is the SetUpAttribute—case-sensitive—and TearDownAttribute that points NUnit at these two methods. It is important to note that each of these methods will be called once before every method attributed with TestAttribute is called.
Next, we need some tests. Tests are adorned with the TestAttribute. Each test is a method that takes no arguments and returns void. Reflection is used to invoke these methods. Technically, return values and parameters can be invoked using Reflection; they would have to be contrived by NUnit because NUnit is running the test. However, it isn't the test method you are testing; it is the code in the test method. Because you write the test code, you can supply the arguments. I defined two positive tests AddIntegerPass and AddDoublePass, on lines 23 through 28, and on lines 31 through 35, respectively. Each of these tests uses the assertion class defined by NUnit. I invoked Assertion.AssertEquals on each of lines 26 and 33. The first argument is a message; the second argument is the value returned by the third argument. For example, line 27 evaluates an Add equal to 15. This should return true. We could also have used 15 as the second argument and the Arithmetic.Add method as the third, rewriting the lines 26 and 27 as Assertion.AssertEquals(message, 15, Arithmetic.Add(10, 5)).
There are several testing tools in the NUnit framework. The online documentation, source, and experimentation will have to be your guide for now. Perhaps because NUnit is open source, some diligent person will contribute help documentation.
Lines 27 through 44 incorporate a second attribute: ExpectedExceptionAttribute. Passing the type record of an exception class, you can test to ensure that a test throws an expected exception. In the example, I intentionally create an overflow condition and want to ensure an exception is thrown.
Finally, the last attribute demonstrated is on lines 46 through 50. The IgnoreAttribute can be used to indicate a test that shouldn't be run by NUnit, perhaps because it isn't quite baked yet.
The way tests are defined are up to you. A great approach is to make testing analysis a deliverable, reviewable aspect of the project. Users can describe good and bad behavior, analysts can record them, and programmers can codify these behaviors. In this manner, testing results are verifiable.