Jesse Johnston
{ ironic tagline here }

FatalExecutionEngineError returning from WPF PageFunction

Friday, 30 November 2007 01:17 by jesse

I'm posting this for future reference.  If you invoke a PageFunction and add an event handler to the PageFunction's Return event, the event handler must be an instance method of the calling page.  No other method will do.  If the event handler doesn't meet this criteria, you'll see the FatalExecutionEngineError thrown when you return from the PageFunction.  Sorry, no separation of concerns allowed.

More details in this forum thread: http://forums.microsoft.com/MSDN/ShowPost.aspx?PostID=1975763&SiteID=1.

kick it on DotNetKicks.com

Digg It!DZone It!StumbleUponTechnoratiRedditDel.icio.usNewsVineFurlBlinkList
Categories:   WPF
Actions:   E-mail | del.icio.us | Permalink | Comments (0) | Comment RSSRSS comment feed

ClickOnce Error after upgrading to VS 2008

Thursday, 29 November 2007 22:04 by jesse

We recently upgraded to VS 2008 RTM at work, and shortly after discovered that a new build of our ClickOnce-deployed WinForms application failed to install correctly!  During the download process, a message box would display, and the log details revealed an InvalidDeploymentException (RefDefValidation) in System.Deployment.Application.DownloadManager.ProcessDownloadedFile().

Let me just say that days of fruitless investigation followed.

I'm happy to report that I finally did find the solution (completely by accident).  A new option has been added to Visual Studio, allowing you to automatically create and embed a manifest in your application.  This manifest allows you to provide proper application behavior when running under UAC in Vista.  Here's the option, on the Application property page of the Visual Studio project:Manifest Options

When you upgrade a WinForms project from VS2005 to VS2008, this option is automatically set for you.  If you had identified the application as a ClickOnce application in VS 2005, the resulting deployment from a VS2008 build works correctly.

Real-world ClickOnce

If you're like me, you're developing in a professional environment where you can't just click the Publish Now button and send an app from your development box into production.  Is it just me, or does everyone think that's insane?  Do you think we might want to QA those phasers first, Captain?

We have a relatively complex build that builds our ClickOnce manifests based on the target environment (Dev, QA, or Production) so that we can test ClickOnce updates from previous versions before going to production.  Also, since we've had our ClickOnce app deployed for more than one year, we had to overcome the certificate renewal problem, which requires a special signing process using both the expired and the current signing certificate.  Needless to say, neither of these can be set up in Visual Studio.  What we ended up with is a build where the ClickOnce options are not set in the Visual Studio project, but externally in an MSBuild project file, using many of the built-in targets that VS uses.

Migrating from VS2005

When the project was converted to VS2008, it was upgraded by the wizard as if it wasn't a ClickOnce project.  This means that the manifest setting was set incorrectly.

Here's what the Visual Studio documentation has to say about the manifest setting:

 

Manifest
Selects a manifest generation option when the application runs on Windows Vista under User Account Control (UAC). This option can have the following values:
  • Embed manifest with default settings. Supports the typical manner in which Visual Studio operates on Windows Vista, which is to embed security information in the application's executable file, specifying that requestedExecutionLevel be AsInvoker. This is the default option.
  • Create application without a manifest. This method is known as virtualization. Use this option for compatibility with earlier applications.
  • Properties\app.manifest. This option is required for applications deployed by ClickOnce or Registration-Free COM. If you publish an application by using ClickOnce deployment, Manifest is automatically set to this option.

The only options I see in the combo box are the first and second.  In our application, "Embed manifest with default settings" was selected by the conversion.  As soon as I chose "Create application without a manifest", I was able to build, deploy, and download with no trouble at all.

I still need to investigate the manifest issue further.  I want my users to have a great UAC experience with our app, and that means getting the manifest right.  For now though, "create without a manifest" is your quick fix.

Click On!

kick it on DotNetKicks.com

Digg It!DZone It!StumbleUponTechnoratiRedditDel.icio.usNewsVineFurlBlinkList
Tags:  
Categories:   .NET
Actions:   E-mail | del.icio.us | Permalink | Comments (0) | Comment RSSRSS comment feed

Service Locator in Code?

Wednesday, 28 November 2007 03:04 by jesse

Phil Haack posed a question today about dependency injection frameworks:

Normally, I would just specify a type to instantiate as another PluginFamily entry. But what I really want to happen in this case is for StructureMap to call a method or delegate and use the value returned as the constructor argument.

...

Does anyone know if something like this is possible with any of the Dependency Injection frameworks out there? Whether via code or configuration?

I don't have much experience with the big three DI frameworks Castle Windsor, Spring.NET and StructureMap, but what I've seen suggests that these are strongly configuration based.

That means that I can specify a concrete type name in a configuration file that the framework will use to instantiate a particular interface type.  For example, in Spring.NET, you might specify the concrete type like this:

<objects xmlns="http://www.springframework.net">
    <object
        name="TheClassIWant"
        type="MyNamespace.MyConcreteClass, MyAssembly"
    </object>
</objects>

Then in your code, you can ask Spring to create an instance of MyConcreteClass like so:

IApplicationContext ctx = ContextRegistry.GetContext();
MyConcreteClass instance = (MyConcreteClass)ctx.GetObject("TheClassIWant");

That's all well and good.  In Phil's case, though, he wants to specify a way to create an instance of MyConcreteClass in code rather than configuration.  Let's take that requirement more generally and assert that we know what concrete classes we want to instantiate in our application, but just don't want the whole application to know.  Maybe I need an instance of ISomeInterface in some part of my code, but that code doesn't know how to get an ISomeInterface.  What I want is some global mechanism to define how to get an instance of some type, and then a way to get the type.  Now, how do we do that in code?

Simple Service Locator

I had some spare time while at DevConnections, and decided to create just that - a code-based service locator.  The basic idea is that we'll have a ServiceLocator class that maps interface types to concrete types.

At some point in your code, you register a concrete type for an interface:

ServiceLocator locator = new ServiceLocator();
locator.Register<ITest>(typeof(Test));

This means that when we ask for instance of the interface ITest, ServiceLocator will create and return an instance of Test (the implication of course is that Test implements ITest).  Here's how we ask for an ITest:

ITest instance = locator.Get<ITest>();

If you expose a singleton instance of ServiceLocator in your application, you can register your types at startup, and then use the locator to instantiate the concrete types from anywhere in the code.

What about parameterized constructors?

Default constructors are the trivial case.  How do we handle creation of concrete types that require constructor parameters?  We have two cases:

  1. The parameter is an interface that we want ServiceLocator to instantiate
  2. The parameter is a value that we want to provide explicitly.

Here's how we handle the first case:

locator.Register<ICat>(typeof(Cat), typeof(IOwner));
locator.Register<IOwner>(typeof(Owner));
 
ICat instance = locator.Get<ICat>();

Here we're saying that if we want an ICat, we need to create a Cat with the constructor Cat(IOwner).  By registering IOwner to create an Owner, ServiceLocator is able to create a Cat by first instantiating an Owner and passing that value to the Cat constructor.

In the second case, we have some value that ServiceLocator doesn't know how to create.  Let's suppose the Cat constructor takes an IOwner and a string:

locator.Register<IOwner>(typeof(Owner));
locator.Register<ICat>(typeof(Cat), new TypeParameter(typeof(IOwner)), new TypeParameter(typeof(string), "Kitty"));
ICat cat = locator.Get<ICat>();

In this case, ServiceLocator will call the constructor Cat(IOwner, string) with an IOwner it creates, and the string "Kitty".

And Singletons?

Sometimes we always want the same instance of an object:

Cat kitty = new Cat();
locator.Register<ICat>(kitty);
ICat cat = locator.Get<ICat>();

Here we always return the instance kitty when an ICat is requested.  We all know how hard it is to work with a large number of cats.

Le Delegate

Yes, there is a Santa Claus.  I might have some code that is responsible for creating an object.  Perhaps it's responsible for managing a pool of objects, or just has some complexity that would be difficult to express in configuration.  I want to register a delegate that will provide the instance.

locator.Register<ICat>(delegate() { return kitty.IsSleeping ? buffy : kitty; });
ICat cat = locator.Get<ICat>();

Here I provided an anonymous method in the Register() call, but I could just as easily register some other class method.

So there you have it - a configuration-free service locator with quite a bit of flexibility in object creation.

You can download the source for my ServiceLocator here.  The project was built in Visual Studio 2008.

Should I Use it?

If it works for you!  It's free of course.  However, I conjured this up in an hour or two, so it's not likely that ServiceLocator offers anywhere near the capabilities of a full-fledged dependency injection framework.  I'd encourage you to look closely at Castle Windsor, Spring.NET and StructureMap.  I'm sure that there are other great frameworks too.

I'd be grateful for any feedback as always.

kick it on DotNetKicks.com

Digg It!DZone It!StumbleUponTechnoratiRedditDel.icio.usNewsVineFurlBlinkList
Categories:   .NET
Actions:   E-mail | del.icio.us | Permalink | Comments (0) | Comment RSSRSS comment feed