C#

Using two test frameworks - NUnit & MSTest

Daniel Gray — Wed, 10 Sep 2008 09:44:17 GMT

Following on from Neil's post about switching between unit test frameworks, I was hitting some problems with his described method. The answer (for me at least, your mileage may vary) was to change around the using block to :

   1:  #if NUNIT

   2:  using NUnit.Framework;

   3:  using TestClass = NUnit.Framework.TestFixtureAttribute;

   4:  using TestMethod = NUnit.Framework.TestAttribute;

   5:  using TestInitialize = NUnit.Framework.SetUpAttribute;

   6:  using TestCleanup = NUnit.Framework.TearDownAttribute;

   7:  using ClassInitialize = NUnit.Framework.TestFixtureSetUpAttribute;

   8:  using ClassCleanup = NUnit.Framework.TestFixtureTearDownAttribute;

   9:  #else

  10:  using Microsoft.VisualStudio.TestTools.UnitTesting;

  11:  #endif

The NUNIT symbol is only defined on debug builds, so I'm using NUnit to run my tests when I'm coding, and then MSTest for builds conducted during CI / Release builds. This gives me the perfect duality of fast turn around of tests whilst coding (including automatic test runs when I build) and the integration of MSTest into TFS during "proper" builds. As I'm using MSTest for release builds I can also enable code coverage (occasionally running the tests through MSTest in development to get a view of how much I've covered).

One thing to watch out for when using Rhino Mock with code coverage is a trust issue, neatly described in this blog entry.

Rhino Mock - Same interface, different types - Revisited

Daniel Gray — Wed, 10 Sep 2008 08:46:07 GMT

Following on from yesterdays post about creating two different types from a single interface - I have my answer. In fact I've got two... one's dirty, one's clean. Thanks to Hadi for the clean answer, which is what I've gone with.

Dirty Answer

Those of a sensitive disposition may want to look away now, or at least scroll to the next paragraph, because this isn't going to be pretty. I dug into the Rhino Mock assembly looking for the point where, maybe, I'd be able to switch the default behaviour to want I wanted. In my quest I found this method on MockRepository:

   1:  protected virtual ProxyGenerator GetProxyGenerator(Type type)

   2:  {

   3:      if (!generatorMap.ContainsKey(type))

   4:      {

   5:          generatorMap[type] = new ProxyGenerator();

   6:      }

   7:      return generatorMap[type];

   8:  }

"Oh yes!" thinks I. All I have to do is inherit MockRepository as MultiTypeMockRepository, override GetProxyGenerator and then always return a new ProxyGenerator. Alas, this was not to be, somehow the people behind Rhino have managed to expose an internal type (ProxyGenerator) as a return value on a protected method of a public class. For a brief moment I contemplated grabbing the Rhino source and rolling my own, but the overhead for a few simple tests wasn't worth it. Finding this method had exposed to me the cache of ProxyGenerators that Rhino are using. My next line of thought was to use reflection to tinker with the innards of MockRepository (told you this was the dirty answer):

   1:     public static class MyMockRepository

   2:     {

   3:        public static void FlushProxyGenerators()

   4:        {

   5:           Type mrType = typeof(MockRepository);

6:

   7:           FieldInfo generatorMapField = mrType.GetField("generatorMap", BindingFlags.NonPublic | BindingFlags.Static);

8:

   9:           object generatorMap = generatorMapField.GetValue(null);

10:

  11:           lock (generatorMap)

  12:           {

  13:              Type generatorMapType = generatorMap.GetType();

  14:              PropertyInfo keys = generatorMapType.GetProperty("Keys");

  15:              MethodInfo remove = generatorMapType.GetMethod("Remove");

16:

  17:              List<Type> types = new List<Type>(keys.GetValue(generatorMap, null) as IEnumerable<Type>);

18:

  19:              foreach (Type t in types) remove.Invoke(generatorMap, new object[] { t });

  20:           }

  21:        }

  22:     }

Guess what? It worked. Although my unit test code now had a rather odd call to FlushProxyGenerators in between the creation of the two outbound classes. That and I'd broken pretty much every rule in the book... dang.

Clean Answer

Hadi had suggested that I create a couple of stub concrete classes from my interface, set the SendMessage method as virtual and then mock from them. The resultant code looks thus:

   1:     public class MockEmailService : IOutboundService

   2:     {

   3:        #region IOutboundService Members

4:

   5:        public virtual void SendMessage(Dictionary<string, string> data, IEnumerable<IServicePolicy> policies)

   6:        {

   7:           throw new NotImplementedException();

   8:        }

9:

  10:        #endregion

  11:     }

   1:     public class MockFtpService : IOutboundService, IInboundService

   2:     {

   3:        #region IOutboundService Members

4:

   5:        public virtual void SendMessage(Dictionary<string, string> data, IEnumerable<IServicePolicy> policies)

   6:        {

   7:           throw new NotImplementedException();

   8:        }

9:

  10:        #endregion

11:

  12:        #region IInboundService Members

13:

  14:        public event EventHandler<OrderReceivedEventArgs> OrderReceived;

15:

  16:        public event EventHandler<OrderNotificationReceivedEventArgs> OrderNotificationReceived;

17:

  18:        public event EventHandler<OrderAssociationReceivedEventArgs> OrderAssociationReceived;

19:

  20:        public event EventHandler<OrderActivationReceivedEventArgs> OrderActivationReceived;

21:

  22:        #endregion

  23:     }

   1:        [TestMethod]

   2:        public void SendMessage_TwoOutbound_RoutedToTwoCorrectly()

   3:        {

   4:           IOutboundService outbound1 = _mockRepository.StrictMock<MockEmailService>();

   5:           IOutboundService outbound2 = _mockRepository.StrictMock<MockFtpService>();

   6:           ProcessOutboundRouter router = new ProcessOutboundRouter();

   7:           router.Add(outbound1);

   8:           router.Add(outbound2);

9:

  10:           IServicePolicy policy1 = _mockRepository.StrictMock<IServicePolicy>();

  11:           IServicePolicy policy2 = _mockRepository.StrictMock<IServicePolicy>();

12:

  13:           Dictionary<string, string> data = new Dictionary<string, string>();

14:

  15:           Expect.Call(policy1.ServiceType).Return(outbound1.GetType());

  16:           Expect.Call(delegate { outbound1.SendMessage(data, new IServicePolicy[] { policy1 }); });

  17:           Expect.Call(policy2.ServiceType).Return(outbound2.GetType());

  18:           Expect.Call(delegate { outbound2.SendMessage(data, new IServicePolicy[] { policy2 }); });

19:

  20:           _mockRepository.ReplayAll();

  21:           router.SendMessage(data, new IServicePolicy[] { policy1, policy2 });

  22:           _mockRepository.VerifyAll();

  23:        }

Note : I've changed the name of some of the interfaces.

Much neater in the test, and without the pangs of guilt I get when I do something really nasty.

Rhino Mock - Same interface, different types

Daniel Gray — Tue, 09 Sep 2008 13:10:05 GMT

I'm trying out Rhino mock on a new project at the moment, trying to get into the TDD groove (which, thanks to NUnit, I am). I've started my codebase with interfaces and a couple of implementations, writing my tests before the classes like a good boy. All was going well, Rhino mock proving to be adaptable and reasonably concise in its operations when I hit upon a scenario that had me stumped.

I have an ProcessOutboundRouter class that implements the interface IProcessOutbound, this router wraps up calls to a list of other IProcessOutbound instances, forwarding messages to the correct instance on the basis of policies (defined by the IServicePolicy interface). The router does this by matching the value of a ServiceType property on the policy to the type of the IProcessOutbound instance. When it came to testing I'd originally setup my test thus:

   1:  [Test]

   2:  public void SendMessage_TwoOutbound_RoutedToTwoCorrectly()

   3:  {

   4:     IProcessOutbound outbound1 = _mockRepository.StrictMock<IProcessOutbound>();

   5:     IProcessOutbound outbound2 = _mockRepository.StrictMock<IProcessOutbound>();

   6:     ProcessOutboundRouter router = new ProcessOutboundRouter();

   7:     router.Add(outbound1);

   8:     router.Add(outbound2);

9:

  10:     IServicePolicy policy1 = _mockRepository.StrictMock<IServicePolicy>();

  11:     IServicePolicy policy2 = _mockRepository.StrictMock<IServicePolicy>();

12:

  13:     Dictionary<string, string> data = new Dictionary<string, string>();

14:

  15:     Expect.Call(policy1.ServiceType).Return(outbound1.GetType());

  16:     Expect.Call(delegate { outbound1.SendMessage(data, new IServicePolicy[] { policy1 }); });

  17:     Expect.Call(policy2.ServiceType).Return(outbound2.GetType());

  18:     Expect.Call(delegate { outbound2.SendMessage(data, new IServicePolicy[] { policy2 }); });

19:

  20:     _mockRepository.ReplayAll();

  21:     router.SendMessage(data, new IServicePolicy[] { policy1, policy2 });

  22:     _mockRepository.VerifyAll();

  23:  }

Alas, 'twas not to be... I'd hoped that the StrictMock<T> method would return different mocks with different types when creating multiple mocks of the same interface. Understandably it doesn't, both outbound1 and outbound2 are of the same proxy type generated by Rhino mock. Creating a new MockRepository instance doesn't help either. So I've, somewhat unfortunately, ended up with the following:

   1:  [Test]

   2:  public void SendMessage_TwoOutbound_RoutedToTwoCorrectly()

   3:  {

   4:     IProcessOutbound outbound1 = _mockRepository.StrictMock<IProcessOutbound>();

   5:     IProcessOutbound outbound2 = _mockRepository.StrictMock<IMockProcessOutbound>();

   6:     ProcessOutboundRouter router = new ProcessOutboundRouter();

   7:     router.Add(outbound1);

   8:     router.Add(outbound2);

9:

  10:     IServicePolicy policy1 = _mockRepository.StrictMock<IServicePolicy>();

  11:     IServicePolicy policy2 = _mockRepository.StrictMock<IServicePolicy>();

12:

  13:     Dictionary<string, string> data = new Dictionary<string, string>();

14:

  15:     Expect.Call(policy1.ServiceType).Return(outbound1.GetType());

  16:     Expect.Call(delegate { outbound1.SendMessage(data, new IServicePolicy[] { policy1 }); });

  17:     Expect.Call(policy2.ServiceType).Return(outbound2.GetType());

  18:     Expect.Call(delegate { outbound2.SendMessage(data, new IServicePolicy[] { policy2 }); });

19:

  20:     _mockRepository.ReplayAll();

  21:     router.SendMessage(data, new IServicePolicy[] { policy1, policy2 });

  22:     _mockRepository.VerifyAll();

  23:  }

Notice the second IProcessOutbound instance is a mock for a different interface (IMockProcessOutbound, which merely inherits IProcessOutbound without adding anything). Whilst this works, it's nowhere near a final solution, what happens if I need a third instance (again of a different type)? Another interface? Not a good pattern.

Has anyone managed to achieve the desired effect without having to resort to the new interface trick?

Extracting public certificates from code signed assemblies

Daniel Gray — Tue, 16 Jan 2007 09:37:00 GMT

Handy method for retrieving the public portion of certificate used in the digital signing of assemblies:

Assembly assm = Assembly.GetAssembly(typeof(CertificateAccess));

string location = assm.Location;

X509Certificate cert = X509Certificate2.CreateFromSignedFile(location);

byte[] raw = cert.GetRawCertData();

X509Certificate2 cert2 = new X509Certificate2();

cert2.Import(raw);

sVerificationCertificate = cert2;

Custom extraction rules for Web Tests

Daniel Gray — Tue, 19 Sep 2006 15:28:00 GMT

To help test MiVoice I've put together a Web Test that votes continously for a random candidate. I tried using the Form Field Extraction rule to select a candidate from the options returned (as a radio button list), but it wasn't up to the job (didn't seem to know what a radio button list was and couldn't randomly select a value). After nosing around in the documentation you can convert a bog standard web test into a coded web test and then from there hook in custom extraction rules. Our custom rule looks like :

using System;
using System.Collections.Generic;
using System.Text;
using Microsoft.VisualStudio.TestTools.WebTesting;
using System.Text.RegularExpressions;
namespace MiVoiceLoadTest
{
    public class GetCandidateOidExtractionRule : ExtractionRule
    {
        private string mFieldName = string.Empty;
        private static Random sRandom = new Random();
        public override string RuleName
        {
            get { return "GetCandidateOidExtractionRule"; }
        }
        public string FieldName
        {
            get { return mFieldName; }
            set { mFieldName = value; }
        }
        public override void Extract(object sender, ExtractionEventArgs e)
        {
            List candidateOids = new List();
            string regexString = "type=\"radio\" name=\"" + FieldName.Replace("$", "\\$") + "\" value=\"([a-z0-9\\-]{36})\"";
            Regex r = new Regex(regexString);
            MatchCollection mc = r.Matches(e.Response.BodyString);
            foreach (Match m in mc)
            {
                string v = m.Groups[1].Value;
                if (!string.IsNullOrEmpty(v))
                {
                    candidateOids.Add(v);
                }
            }
            int i = sRandom.Next(candidateOids.Count);
            string oid = candidateOids[i];
            e.WebTest.Context.Add("CandidateOid", oid);
            e.Success = true;
            e.Message = "Passed";
        }
    }
}

Notice the e.WebTest.Context.Add("CandidateOid"), this puts the value we selected into the context for the next request to use. This rule is then hooked into the request that's going to return the radio button list as

GetCandidateOidExtractionRule candidates = new GetCandidateOidExtractionRule();
candidates.FieldName = "ctl00$pageContent$vpVote$wizVotingProcess$rblCandidates";
candidates.ContextParameterName = "1"; // ! THIS BIT IS VITAL, NULLRFERENCEEXCEPTION OTHERWISE
request2.ExtractValues += new EventHandler(candidates.Extract);

Then the value we've set into the context is consumed in a subsequent request thus

request3Body.FormPostParameters.Add("ctl00$pageContent$vpVote$wizVotingProcess$rblCandidates", this.Context["CandidateOid"].ToString());

Worked out very nicely.

lock(typeof(MyLockedType)) is bad, mmmm'ok?

Daniel Gray — Wed, 02 Aug 2006 08:47:00 GMT

Seems that MS have been promoting bad practice in some of their documentation...

http://msdn.microsoft.com/library/default.asp?url=/archive/en-us/dnaraskdr/html/askgui06032003.asp

Long and the short is don't lock on a type, so:

public class MyLockedType
{
  public static void DoWork()
  {
     lock(typeof(MyLockedType))
     {
       ...do something here
     }
  }
}

is bad and should be rewritten as:

public class MyLockedType
{
  private static object mLockObj = new object();

  public static void DoWork()
  {
     lock(mLockObj)
     {
       ...do something here
     }
  }
}

There's more details in the article, but the main problem boils down to not having control over the MyLockedType to the extent that code outside of the class can lock on the type. Not only does this break the encapsulation but could lead to some very fun debugging.

ZedGraph - open source charting component

Daniel Gray — Mon, 24 Jul 2006 14:38:00 GMT

Just started using a rather nifty little charting component at home by the name of ZedGraph. It's released under the LGPL which means it can be linked to without "infecting" commerical apps with the license (one of the criticisms oft leveled at the GPL). There's a handy set of tutorials / examples available online and it works better than a lot of commercial components I've tried (built in panning / zooming / save as image, etc).

It's .NET 2.0 only unfortunately but is capable of some fairly impressive visuals (for something that's free) and runs in both WinForms and ASP.NET.

Tracing between AppDomains

Daniel Gray — Thu, 20 Jul 2006 13:22:00 GMT

Problem :

AppDomain 1 creates a new AppDomain (2) to run tasks within, tracing from AppDomain 2 needs to be conducted in AppDomain 1 to allow external app (NUnit) to access the Console where it's all being written.

Solution :

4 new classes :

RemoteAppDomainEventHandlers - serializable class containing the event handling code to run in AppDomain 1
RemoteTraceEventArgs - serializable event args used to transport tracing info between AppDomains
RemoteTraceListener - sub class of TraceListener, wraps a RemoteTracing instance
RemoteTracing - MarshalByRefObject that receives trace messages from RemoteTraceListener and raises events back to AppDomain 1

When AppDomain 2 is created an instance of RemoteTracing is created and unwrapped into AppDomain 1, EstablishRemoteTracing is called which sets up a RemoteTraceListener in AppDomain 2, finally the trace event is wired up to the RemoteTrace handler on an instance of RemoteAppDomainEventHandlers.

// we need to hook in tracing from the app domain back to here
m_remoteTracing = (RemoteTracing)m_appDomain.CreateInstanceAndUnwrap("com.iMeta.ApplicationUpdate","com.iMeta.ApplicationUpdate.Remote.RemoteTracing");
m_remoteTracing.EstablishRemoteTracing();
m_remoteTracing.Trace+=new TraceEventHandler(m_eventHandlers.RemoteTrace);

public void EstablishRemoteTracing()
{
   // clear any existing trace listeners
   System.Diagnostics.Trace.Listeners.Clear();
   // create one of our remote listeners that wraps this RemoteTracing object
   RemoteTraceListener rtl = new RemoteTraceListener(this);
   System.Diagnostics.Trace.Listeners.Add(rtl);
}

Then when a call is made to Trace.Write in AppDomain 2 it hits the following code in the RemoteTraceListener

public override void Write(string message, string category)
{
   m_remote.SendTraceToClient(message,category);
}

Which in turn calls the RemoteTracing instance it was created with:

public void SendTraceToClient(string message, string category)
{
   // create a RemoteTraceEventArgs object and raise event with it
   RemoteTraceEventArgs rtea = new RemoteTraceEventArgs(message,category);
   OnTrace(rtea);
}

An event gets raised in AppDomain 2 and AppDomain 1 handles it by writing the trace in it's domain.

Originally the event handler for AppDomain 1 was on the class that loaded the AppDomain, this didn't seem to work too well as the event handlers were running in AppDomain 2 which caused some fun StackOverflowExceptions!

Forcing metaSite references from a new metaCore model

Daniel Gray — Wed, 31 May 2006 10:30:00 GMT

If you're using the new metaCore/metaSite you'll need to reference the metaSite.Core assembly from your metaCore model. But the compiler will helpfully remove any references it considers redundant, so you'll need to add a piece of code to force the compiler to include a reference to metaSite. We had this early on in MiVoice - Vote, but have just had a resurfacing of the issue when doing a release build. Previously we had the following included in the model assembly:

using System;
using System.Collections.Generic;
using System.Text;

namespace com.iMeta.MiVoice.Vote
{
   static class metaSiteReferenceForce
   {
      private static void Nothing()
      {
         com.iMeta.metaSite.schema.msURL m = null;
      }
   }
}

which worked fine for debug builds. Having been fighting with release builds we hit an issue where the compiler does further optimisations and completely removed the null reference above (very cleverly). So we've ended up with:

using System;
using System.Collections.Generic;
using System.Text;

namespace com.iMeta.MiVoice.Vote
{
   static class metaSiteReferenceForce
   {
      private static void Nothing()
      {
         com.iMeta.metaSite.schema.msURL m = com.iMeta.metaSite.schema.msURL.Factory.Create();
      }
   }
}