dotnet, gists, aspnet, csharp comments edit

With Autofac (particularly the multitenant extensions) I see a lot of questions come through that boil down to this:

I have an ASP.NET MVC project. I have some controllers in a plugin assembly that isn’t referenced by the main project. At application startup, I do some scanning and use Autofac to dynamically register the controllers. For some reason I get an error when I try to visit one of these controllers. How can I have a controller in a plugin assembly?

Shannon Deminick has a nice blog article that explains this in a different context - similar question, but the same answer.

The problem is that the default controller factory in ASP.NET MVC 3 and 4 (the latest version as of this writing) is really tied to the BuildManager. The BuildManager is the class that maintains the internal list of all the referenced application assemblies and handles ASP.NET compilation.

The DefaultControllerFactory in ASP.NET MVC, in the CreateController method, uses a ControllerTypeCache internal type to locate the controller type being instantiated. The ControllerTypeCache uses another internal, TypeCacheUtil, to load the set of controllers from the list of referenced assemblies. TypeCacheUtil uses the BuildManager.GetReferencedAssemblies() method to initialize that list. BuildManager.GetReferencedAssemblies() includes:

  • Assemblies that are referenced by the main web application.
  • Assemblies you list in the <assemblies> part of web.config.
  • Assemblies built from App_Code.

Note that none of those automatically include non-referenced, already-built plugin assemblies.

You need to add your plugin assembly to the list of referenced assemblies in the BuildManager.

You can do that in one of three ways.

First, you can register the assembly in web.config. This makes for a less “drop-in-an-assembly” plugin model, but it also means no code getting executed. If you put your plugins in a folder other than bin, you will also have to register that path. Here’s a snippet showing a web.config with this sort of registration.

<?xml version="1.0" encoding="utf-8"?>
<configuration>
  <runtime>
    <assemblyBinding xmlns="urn:schemas-microsoft-com:asm.v1">
      <!--
          IF you put your plugin in a folder that isn't bin, add it to the probing path
      -->
      <probing privatePath="bin;bin\plugins" />
    </assemblyBinding>
  </runtime>
  <system.web>
    <compilation>
      <assemblies>
        <add assembly="The.Name.Of.Your.Plugin.Assembly.Here" />
      </assemblies>
    </compilation>
  </system.web>
</configuration>

Another alternative is to register your assemblies in code at PreApplicationStart. This is the method outlined in Deminick’s article in more detail. The idea is that you use the PreApplicationStartMethodAttribute to bootstrap some assembly scanning and registration with the BuildManager (since that all has to happen before app startup).

Basically, you mark your main project assembly with the attribute and point to a class that has a static method that will do the auto-registration, like this:

[assembly: PreApplicationStartMethod(typeof(Initializer), "Initialize")]

Then you write your initializer class to do the assembly loading and registration with the BuildManager. Something like this:

using System.IO;
using System.Reflection;
using System.Web.Compilation;

namespace MyNamespace
{
  public static class Initializer
  {
    public static void Initialize()
    {
      var pluginFolder = new DirectoryInfo(HostingEnvironment.MapPath("~/plugins"));
      var pluginAssemblies = pluginFolder.GetFiles("*.dll", SearchOption.AllDirectories);
      foreach (var pluginAssemblyFile in pluginAssemblyFiles)
      {
        var asm = Assembly.LoadFrom(pluginAssemblyFile.FullName);
        BuildManager.AddReferencedAssembly(asm);
      }
    }
  }
}

A third way would be to create your own ControllerFactory implementation. In your custom controller factory you could search your plugin assemblies for the controller types or use some other convention to determine which controller type to resolve. I don’t have any sample code for that and there is a lot of work to accomplish that and get it right - supporting areas, properly handling the type resolution… If you go that route, and some people have, you’ll have to go out searching for samples. I don’t have any here to readily provide.

I’d recommend one of the first two options. They’re the easiest and require the least “messing around with the framework” to get things working for you.

General Ramblings comments edit

Jeff Atwood just posted an article about Aaron Swartz, his unfortunate story, and the notion of ragequitting.

I agree with Jeff on Swartz and the thoughts about that case. Rather than restate all that, check out the article. My thoughts go out to Swartz and his family. He’ll be missed.

What I disagree with is this:

Ragequitting is childish, a sign of immaturity.

We’ve often used “vote with your feet” as an idiom describing how people can effectively support (or show lack of support) for a thing, particularly an internet (or programming, or technology) thing. It occurs to me that ragequitting, while abrupt, is effectively foot-voting-in-action. I’ve done it myself, I admit, and I’m sure you have, too. Point is, just because it’s fast or unannounced doesn’t mean it’s any less valid, and, in my opinion, certainly doesn’t mean it’s childish. It’s within everyone’s rights to choose their situation and decide what’s best for them regardless of the emotion that may be associated with said decision.

windows comments edit

I swear every time I change the Java settings to stop auto-updating it still pops up that stupid “A Java Update is Available” toast in the corner and I want to punch it repeatedly. Killing the scheduled task from startup works until you actually do install the next update, at which point you forget it and it puts itself back.

I run as a non-admin user. The Java auto-update thing hates that. It tells me there’s an update, then I say, “OK, do it then.” It asks me for admin credentials, I enter them, and I instantly get a failure message. Again, I want to punch it repeatedly.

The only way I can get this thing to go away is to manually run the update (or download the entire package and manually install the update). For my own reference, here’s how I do it:

  1. Log in as the administrative user.
  2. Run “C:\Program Files (x86)\Common Files\Java\Java Update\jucheck.exe” with elevated privileges.
  3. Follow the prompts to get the update and make sure to uncheck all the freeware crap they want to install alongside it.

dotnet, gists, build, teamcity comments edit

We use TeamCity as our build server and one of the cool things TeamCity has built in is the ability to serve as a NuGet server. You build your product, run a nuget packtask on it, and TeamCity will automatically add it to the feed.

One of the somewhat odd things I’ve found with TeamCity’s NuGet server is that it seems to require that you let TeamCity run the actual nuget packon packages it should host. That is, even if you wanted to do that action in your build script, you can’t – simply putting the package into your set of build artifacts doesn’t get it into the feed. You actually have to use the “NuGet Pack” build step in your build. When you do that, the build step ignores any version information you put inside your .nuspec files because the “NuGet Pack” build step requires you to specify the version right there.

That’s fine as long as the build number for the build (or some similar existing variable) is also the version you want on your NuGet package. But when you want to have tighter control over it, like calculating the version as part of a build task, it becomes less clear how to get things working. This should help you.

First, you have to establish a NuGet package configuration variable. You need this so you can use it in the NuGet Pack build steps. In your TeamCity build configuration, go to the “Build Parameters” tab and define a “System Property” with your calculated NuGet package semantic version. I called mine “CalculatedSemanticVersion” so it ends up showing in the TeamCity UI as “system.CalculatedSemanticVersion” like this:

system.CalculatedSemanticVersion = 0.0.0 Set it to some simple, default value. It won’t stay that value so it doesn’t matter; it’s more for when you come back later and look at your configuration – this way it’ll make a little more sense.

Next, set up your NuGet Pack build steps. Use this new “system.CalculatedSemanticVersion” property as the NuGet package version you’re building.

On the NuGet Pack step use the new version variable.

Finally, insert a build script step before all of your NuGet Pack steps. In that build script step, calculate the version you really want for your packages and use a TeamCity message to update the variable value. You do that by using a specially formatted message written to the console, like this:

##teamcity[setParameter name='system.CalculatedSemanticVersion' value='1.0.0-beta1']

In MSBuild, you might have something that looks like this:

<?xml version="1.0" encoding="utf-8"?>
<Project
  DefaultTargets="SetVersion"
  xmlns="http://schemas.microsoft.com/developer/msbuild/2003"
  ToolsVersion="4.0">
  <Target Name="SetVersion">
    <!--
      Calculate your semantic version however you like.
      This example uses a made-up build task, but you
      could do anything.
    -->
    <CalculateMySemanticVersion>
      <Output TaskParameter="Version" PropertyName="SemanticVersion" />
    </CalculateMySemanticVersion>
    <!-- The message task here is the important part. -->
    <Message Text="##teamcity[setParameter name='system.CalculatedSemanticVersion' value='$(SemanticVersion)']" />
  </Target>
</Project>

Now when your build configuration runs, the script will calculate your NuGet package version and update the value of the property before the NuGet Pack tasks run. The NuGet Pack tasks will build your packages using the correct calculated version that you controlled through script.

dotnet, gists, build, xml comments edit

Although I wasn’t a big fan of NuGet when it started getting big, I have to admit it’s grown on me. I think part of that has to do with the large amount of improvement we’ve seen since back then. Regardless, I’m in a position with Autofac and other projects where I’m not only consuming NuGet packages, I’m also producing them.

One of the biggest pains I have when maintaining the .nuspec files for my packages is that you can update a dependency for your project (via NuGet) but the corresponding version value isn’t updated in the .nuspec. (This is, of course, assuming you’re doing manual maintenance and not re-generating everything each time. In a single-package solution, I can see regenerating would be fine, but when you’ve got multiple like in Autofac, you don’t want to regenerate.)

What I want is for the .nuspec file <dependency> entries to match the installed package versions that I’m actually building against.

So… I automated that with MSBuild. Here’s how:

First, put placeholders into your .nuspec file(s) using a special format, like this:

<dependencies>
  <dependency id="Autofac" version="$version_Autofac$" />
  <dependency id="Castle.Core" version="$version_Castle.Core$" />
</dependencies>

Each dependency gets a $version_NuGetPackageName$ format placeholder. The “NuGetPackageName” part matches the name of the dependency (and, coincidentally, the first part of the folder name under “packages” where the dependency gets installed in your solution).

Next, in your build script, include a custom task that looks like this. It will look in the “packages” folder and parse the various folder names into these placeholders so you can do some search-and-replace action.

<UsingTask TaskName="GetNuGetDependencyVersions" TaskFactory="CodeTaskFactory" AssemblyFile="$(MSBuildToolsPath)\Microsoft.Build.Tasks.v4.0.dll">
  <ParameterGroup>
    <PackageInstallDirectory Required="true" />
    <Dependencies ParameterType="Microsoft.Build.Framework.ITaskItem[]" Output="true" />
  </ParameterGroup>
  <Task>
    <Using Namespace="System" />
    <Using Namespace="System.Collections.Generic" />
    <Using Namespace="System.IO" />
    <Using Namespace="System.Text.RegularExpressions" />
    <Using Namespace="Microsoft.Build.Framework" />
    <Using Namespace="Microsoft.Build.Utilities" />
    <Code Type="Fragment" Language="cs">
      <![CDATA[
        // Match package folders like Castle.Core.3.0.0.4001
        // Groups[1] = "Castle.Core"
        // Groups[2] = "3.0.0.4001"
        var re = new Regex(@"^(.+?)\.(([0-9]+)[A-Za-z0-9\.\-]*)$");

        try
        {
          // Create item metadata based on the list of packages found
          // in the PackageInstallDirectory. Item identities will be
          // the name of the package ("Castle.Core") and they'll have
          // a "Version" metadata item with the package version.
          var returnItems = new List<ITaskItem>();
          foreach(var directory in Directory.EnumerateDirectories(PackageInstallDirectory))
          {
            var directoryName = Path.GetFileName(directory);
            var match = re.Match(directoryName);
            if(!match.Success)
            {
              continue;
            }
            var name = match.Groups[1].Value;
            var version = match.Groups[2].Value;
            var metadata = new Dictionary<string, string>();
            metadata["Version"] = version;
            var item = new TaskItem(name, metadata);
            returnItems.Add(item);
          }
          Dependencies = returnItems.ToArray();
          return true;
        }
        catch(Exception ex)
        {
          Log.LogErrorFromException(ex);
          return false;
        }
      ]]>
    </Code>
  </Task>
</UsingTask>

If you’re industrious, you could package that build task into an assembly so it’s not inline in your script, but… I didn’t. Plus this lets you see the source.

Now you can use that build task along with the MSBuild Community Tasks “FileUpdate” task to do some smart search and replace. Here are a couple of MSBuild snippets showing how:

<!-- At the top/project level... -->
<!-- You need MSBuild Community Tasks for the FileUpdate task. -->
<Import Project="tasks\MSBuild.Community.Tasks.targets" />
<PropertyGroup>
  <!-- This is where NuGet installed the packages for your solution. -->
  <PackageInstallDirectory>$(MSBuildProjectDirectory)\packages</PackageInstallDirectory>
</PropertyGroup>

<!-- Inside a build target... -->
<ItemGroup>
  <!--
      This should include all of the .nuspec files you want to update. These should
      probably be COPIES in a staging folder rather than the originals so you don't
      modify the actual source code.
  -->
  <NuspecFiles Include="path\to\*.nuspec" />
</ItemGroup>
<!--
    Parse out the installed versions from the list of installed
    packages using the custom task.
-->
<GetNuGetDependencyVersions PackageInstallDirectory="$(PackageInstallDirectory)">
  <Output TaskParameter="Dependencies" ItemName="LocatedDependencies" />
</GetNuGetDependencyVersions>
<!-- Use the MSBuild Community Tasks "FileUpdate" to do the search/replace. -->
<FileUpdate
  Files="@(NuspecFiles)"
  Regex="\$version_%(LocatedDependencies.Identity)\$"
  ReplacementText="%(LocatedDependencies.Version)" />

Generally what you’ll want to do from a process perspective, then, is:

  1. Build and test your project as usual.
  2. Create a temporary folder to stage your NuGet packages. Copy the .nuspec file in along with the built assemblies, etc. in the appropriate folder structure.
  3. Run the file update process outlined above to update the staged .nuspec files.
  4. Run nuget packon the staged packages to build the final output.

This will ensure the final built NuGet packages all have dependencies set to be the same version you’re building and testing against.