Initialize log4Net as early as possible with NUnit

问题:

I'm wondering what is the best way to initialize log4Net in a NUnit project. Of course I want to call the init code (ie. XmlConfigurator.Configure()) as soon as I can to get as many early log output as I can. But since my project is run through NUnit, I have little control on its entry point.

According to NUnit documentation, it should call some constructors first, then a method marked with the [SetUp] attribute in a class marked with [TestFixtureSetup].

So, First, I created a static helper class which I can call several times without trouble.

  public static class LoggingFacility
  {
    private static bool _loggerIsUp = false;

    public static void InitLogger()
    {
      if (_loggerIsUp == false)
        XmlConfigurator.ConfigureAndWatch(f);

      _loggerIsUp = true;
    }
  }

Then, I made all my [TestFixtureSetup] inherit a single one that does pretty much nothing else than calling LoggingFacility.initLogger(). But that still leaves all constructors that are run earlier, in an order I can only assume random. And moreover, it will probably do some static initializations before I am even able to execute some code.

In fact, as I can see in my log, the first 4 seconds or so of execution are completely unrecorded.

Does it mean I will have to call my InitLogger() in every constructor and forbid the use of any static initializer? That's tough job!

Does somebody know a magic trick with this?

回答1:

For single initialization point you should use class marked with [SetUpFixture] attribute and method marked with [SetUp], for example:

[SetUpFixture]
public class TestsInitializer
{
    [SetUp]
    public void InitializeLogger()
    {
        LoggingFacility.InitLogger();
    }
}

Now, this method ([SetUp] InitializeLogger) will run before any test is run, same as one marked with [TearDown] will run once all tests are run. But here's the catch - what does any and all mean in this context? Tests from classes declared in the same namespace as class marked with [SetUpFixture].

For example, assuming hierarchy like this:

- Tests
--- Business
----- TestsInitializer.cs // SetUpFixture class
----- FirstBusinessTests.cs
----- SecondBusinesTests.cs
--- ComplexLogic
----- VeryComplexLogicTests.cs

First and SecondBusinessTests will run after SetUp from TestsInitializer, however VeryComplexLogicTests might run in random order.

According to linked documentation, if you declare SetUpFixture class outside of any namespace, setup and teardown will apply for entire assembly:

Only one SetUpFixture should be created in a given namespace. A SetUpFixture outside of any namespace provides SetUp and TearDown for the entire assembly.

回答2:

A work mate provided me with the following workaround, that does the job:

In all my classes that require logging, I had the following logger initialization

private static readonly ILog Log = LogManager.GetLogger(System.Reflection.MethodBase.GetCurrentMethod().DeclaringType);

I simply changed it to a singleton initializer

private static readonly ILog Log = LoggingFacility.GetLoggerWithInit(System.Reflection.MethodBase.GetCurrentMethod().DeclaringType);

/*** ... ***/

public static class LoggingFacility
{
  private static bool _loggerIsUp = false;

  public static ILog GetLoggerWithInit(Type declaringType)
  {
    if (_loggerIsUp == false)
      XmlConfigurator.Configure(_log4NetCfgFile);
    _loggerIsUp = true;
    return LogManager.GetLogger(declaringType);
  }
}

Because I have this code in every class, this static initializer has to be called very early by NUnit wile instantiating my test classes.

Next step is to make that thread safe :(