CodeCoverage vs ExpectedException

Question

I've got several unittests of this pattern:

[TestMethod ()]
[ExpectedException (typeof (ArgumentNullException))]
public void DoStuffTest_Exception ()
{
    var foo = new Foo ();
    Foo.DoStuff (null);
}

It turns out that code coverage markes the throwing line as half-run, so I get 1 block of uncovered code each time.

After thinking about this problem for a while, the best solution I could come up with was adding a try/catch. Since this is a repeated pattern, I'm going to create a helper method along the lines of

public static void ExpectException<_T> (Action action) where _T: Exception
{
    try { action(); }
    catch (_T) { return; }
    Assert.Fail ("Expected " + _T);
}

This would have the nice side benefit that I could add all exception tests to the non-throwing tests.

Is this a valid design, or did I miss something?

Edit: Ugs... seems like the above ExpectException method leaves me with 1 uncovered block as well.

Answer 1

What you are suggesting is valid. Aside from you code coverage issue, I would argue it is better than using the ExpectedException attribute as it explicitly shows which line of the test is expected to throw the exception. Using ExpectedException means that any line of code in the test can throw the expected exception type and the test will still pass. If the error originates from another call that was not expected to throw, it can disguise the fact that the test should be failing because the line that should be throwing isn't.

What would be a useful modification to what you have proposed would be to return the caught exception:

public static _T ExpectException<_T> (Action action) where _T: Exception
{
    try { action(); }
    catch (_T ex) { return ex; }
    Assert.Fail ("Expected " + typeof(_T));
    return null;
}

This would enable the test code to further assert the exception if it desired (ie. to check a particular message was used).

NUnit (though it doesn't look like you are using it as you have a TestMethod attribute) has a built-in construct similar to what you have proposed:

Assert.Throws<ArgumentNullException>(() => Foo.DoStuff(null))

Answer 2

@adrianbanks the ExpectException does not work as expected if the action parameter throws another exception than the expected exception:

[TestMethod]
public void my_test()
{
    ExpectException<InvalidOperationException>(delegate()
    {
        throw new ArgumentException("hello");
    });
}

When I execute the TestMethod "my_test" i just got a message saying that the test method raised and System.ArgumentException: hello. In this case it should says "Expected InvalidOperationException". I propose a new version for the ExpectException method:

public static void VerifierException<T>(Action action) where T : Exception
{
    try
    {
        action();
    }
    catch (Exception ex)
    {
        Assert.IsInstanceOfType(ex, typeof(T));
        return;
    }

    Assert.Fail("Aucune exception n'a été déclenchée alors qu'une exception du type " + typeof(T).FullName + " était attendue");
}

Answer 3

I know this is an old topic, but I ran into the same problem.

Eventually I questioned myself: why do I need to know the coverage of the tests? I don't! - So let's rule them out, so the coverage is cleaner.

In my test project I've added an CodeCoverage.runsettings file and this is the content:

<?xml version="1.0" encoding="utf-8" ?>
<RunSettings>
  <DataCollectionRunSettings>
    <DataCollectors>
      <DataCollector friendlyName="Code Coverage" uri="datacollector://Microsoft/CodeCoverage/2.0" assemblyQualifiedName="Microsoft.VisualStudio.Coverage.DynamicCoverageDataCollector, Microsoft.VisualStudio.TraceCollector, Version=11.0.0.0, Culture=neutral, PublicKeyToken=b03f5f7f11d50a3a">
        <Configuration>
          <CodeCoverage>
            <ModulePaths>
              <Exclude>
                <ModulePath>.*tests.dll</ModulePath>
                <ModulePath>.*Tests.dll</ModulePath>
                <!-- Add more ModulePath nodes here. -->
              </Exclude>
            </ModulePaths>
          </CodeCoverage>
        </Configuration>
      </DataCollector>
    </DataCollectors>
  </DataCollectionRunSettings>
</RunSettings>

After selecting this Test Settings file my Code Coverage is 100%

This way there is no need to 'hack' the unit test code coverage system, just in order to achieve 100% :-)

Answer 4

Yep this is pretty standard fare - a lot of our tests do the same. At the same time, you have to wonder if you're not placing too high a value on code coverage if those half-branches weigh in so much for it to be worth the effort.