Anymore, if you want to generate documentation, you’re using Sandcastle along with, most likely, some community-generated wrapper for it so you’re not manually executing the bajillion steps Sandcastle requires to get things done. (Wow, that didn’t sound bitter at all.) I use Sandcastle Help File Builder for several projects and like it a lot.

When Sandcastle generates its reflection information (so it can properly resolve links in your docs), you need to tell it where it can find third-party dependencies so it can generate the information it needs. All of this has to happen on the command line (no, there’s not currently a config file option for this), which means when your project gets to be of a significant enough size, you start running into command line length limitations. Sandcastle Help File Builder addresses this by copying all of your dependencies locally into a temporary folder and passing a single dependency location (that folder) to Sandcastle.

That works great… until you run into an issue where you inadvertently rely on multiple versions of a single dependency.

For example - say you have a custom NAnt task project with a compile-time dependency on log4net 1.2.1.40796 and NAnt 0.85.2344.0. NAnt 0.85.2344.0 has a reference to log4net 1.2.9.0. What that means is, unfortunately, that Sandcastle needs access to both versions of the log4net.dll. What it further means is that when Sandcastle Help File Builder tries to copy both versions locally to help you with the dependency thing, there can be only one and Sandcastle dies a miserable death telling you that it can’t find the reference.

How do you resolve that?

Luckily, Sandcastle Help File Builder has a plugin model that allows you to swap in your own behavior for various stages of execution, and that’s what I’ve done.

Since Sandcastle needs the assembly metadata and it doesn’t matter what the physical filename is, I’ve replaced the existing dependency copy routine so that it still copies all of your specified dependencies but it copies them with GUID filenames - so you can have more than one version of a dependency without having a filename clash. Sandcastle resolves the dependencies correctly (it literally just iterates through all of the copied dependencies and caches the metadata) and docs get generated. Peace on Earth is restored.

To use it, just drop it in the PlugIns folder in your Sandcastle Help File Builder installation location. More on using plugins with Sandcastle Help File Builder is included in the SHFB documentation. (Basically, put it in the folder and SHFB will pick it up automatically.)

Grab the compiled version, the source (with unit tests), or both. Free, as usual.

[Download Paraesthesia.SandcastleHelpFileBuilder.Plugin 1.0.0.0 Compiled]

[Download Paraesthesia.SandcastleHelpFileBuilder.Plugin 1.0.0.0 Source]

Version History: 1.0.0.0: First release.

Sunday Jenn, I, and our friends Angela and Keaka gathered up in the Wagon Queen Family Truckster and headed down to Lincoln City, OR. Every year they place a bunch of glass floats up and down the beach, and Angela’s totally into floats, so we went to help her find one.

After a couple of hours walking around on the beach, searching with the other treasure hunters, we came up empty-handed. I was tempted to run to a local shop and just buy a float and then hide it in the sand somewhere to “find,” but I couldn’t sneak away. Somehow I think Angela would have known anyway and probably given me a stern talking-to.

Next time I think I want to blow my own glass float, though it costs like $65 to do it. I think it’d be a pretty neat experience and you’d get to take something cool home. I’ve never blown glass before. I did get myself some Mo’s clam chowder, which is always worth the trip. And now, back to work!

Saturday was my seventh in a series of laser hair removal treatments for my face. I had some excellent results from last time and my facial hair is getting pretty noticeably patchy. Especially at the end of the day, I get sort of “five o’clock dirty-face” now since it’s less “shadow” and more sparse. (I went to the dermatologist the other day and forgot to mention I was getting the treatments. He was all, “I don’t remember you mentioning you had really bad acne when you were younger…” It’s pretty patchy.)

The area around my mouth is being a little stubborn, but my chin is the thickest/coarsest hair on my face, so that’s to be expected, and we hadn’t done the MedioStar laser around my lips because that’s apparently the most painful area. To remedy some of that, this time we did MedioStar around the lips (not actually any more or less painful than anywhere else) and went over my chin twice. We also did full face with the DermoFlash IPL. We should be getting some great results this time, though my face burned for a day like a bad sunburn and it hurts to shave because it’s very tender. My patchiness is also far more visible this morning because I have a small amount of bruising, which is not uncommon and generally goes away in four or five days.

Treatment is getting noticeably less painful every time. I’m pretty sure that doing the DermoFlash for a few treatments was the right way to go (for me) because it thinned things out enough that the MedioStar became bearable. They also fairly recently got this cold air machine that blows supercooled air and freezes you right up, making things, again, more bearable. It beats the crap out of putting ice on your face, which doesn’t work.

The Verizon Fios installers just left, and I’ve got my 15mbps up/2mpbs down running in full form. Got everything transferred over to use the Verizon router and Xbox Live, my primary network problem child, seems to be working swimmingly.

Of course, it didn’t come off without a hitch - the person taking our order mistyped the phone number, so the installers couldn’t really transfer our phone number from Comcast to Verizon because the paperwork didn’t match. So now we have Verizon video and data, and we have both Verizon and Comcast phone. How does that work? Verizon is actually hooked up and running but with the wrong phone number. We’ve had to temporarily keep the Comcast voice service, even if no Comcast junk is hooked up, so we don’t lose our number. It will take a week of futzing around through annoying phone trees and fighting with service people who can’t get off the standard script in order for us to get the number switched. A week to do a two minute job.

Oh, Verizon. I knew you had horrible customer service, and so far, well, I wasn’t wrong.

I like the menus and the TV features a lot better than the Comcast stuff. The UI is far slicker and has a lot of little helpful things in it than the Comcast one did. The net connection seems pretty nice, too, but I can’t really speak to it much since I’ve only written this blog entry and gotten things basically connected. We’ll see how she goes soon.

Oh, and it’s like $50/month cheaper, which doesn’t hurt.

Getting TypeMock, NUnit, and NCover to work together in your build script can sometimes be a tricky thing. Getting any one of those things to work individually is easy enough, getting two going is a little tougher, but getting all three together requires a bit of finesse. Add to that the fact that you may run different versions of different products (different NUnit versions, for example) for different source code bases and it gets downright complicated.

The way my product code works, when I check out the codeline I want to work on it comes with all of the dependencies - every third party assembly, every build tool. That includes TypeMock, NUnit, and NCover. That way the build server doesn’t have to have anything installed when it runs a build - it can auto-deploy TypeMock, register NCover, do its thing, and undo all of that when it’s done. (Yes, there are some drawbacks to that - parallel builds are limited when registered versions of profilers change, for example - but we’ve dealt with that sort of thing in other ways.)

On developer workstations, we have TypeMock installed so we can make use of the tracer and other helpful utilities, but on the build server, we auto-deploy TypeMock.

Since TypeMock is a profiler, if you use it in your unit tests, you can’t just run NUnit in the build and have it work - you have to start TypeMock, then run NUnit, then shut TypeMock back down. If you’re using NCover, you have to make sure NCover is registered and linked with TypeMock.

TypeMock comes with some custom build tasks to help you get this working. You will also want to get NCoverExplorer and NCoverExplorer.Extras to get this working well. NCoverExplorer will aggregate coverage logs for you and NCoverExplorer.Extras comes with NCover and NCoverExplorer MSBuild tasks.

The general flow of what needs to happen is this:

1. Register TypeMock with the system (if it’s not a developer workstation - our devs have it installed).
2. Register NCover with the system.
3. Start TypeMock and link it with NCover.
4. Run NCover and pass it the command line parameters to run your NUnit tests. Tell it which assemblies to profile.
5. Stop TypeMock.
6. Unregister NCover with the system.
7. Use NCoverExplorer to aggregate the coverage reports into a single report.
8. On error, stop TypeMock and unregister NCover.

Here’s the example:

<?xml version="1.0" encoding="utf-8" ?>
<Project xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
  <!-- Register the tasks necessary for running tests and coverage. -->
  <Import Project="Relative\Path\To\TypeMock.NET\TypeMock.MSBuild.Tasks"/>
  <UsingTask
    TaskName="NCover"
    AssemblyFile="Relative\Path\To\NCoverExplorer.Extras\NCoverExplorer.MSBuildTasks.dll"/>
  <UsingTask
    TaskName="NCoverExplorer"
    AssemblyFile="Relative\Path\To\NCoverExplorer.Extras\NCoverExplorer.MSBuildTasks.dll"/>

  <PropertyGroup>
    <!-- Property indicating we're building on a dev machine - build server will set this to false. -->
    <DeveloperBuild>true</DeveloperBuild>
    <!-- Property indicating where build logs will go. -->
    <BuildLogDirectory>Relative\Path\To\Logs</BuildLogDirectory>
  </PropertyGroup>

  <Target Name="test">
    <!-- Register TypeMock only if it's the build server - it'll already be on a developer box. -->
    <TypeMockRegister
      Company="YOUR COMPANY"
      License="YOUR LICENSE"
      AutoDeploy="true"
      Condition="'$(DeveloperBuild)'!='true'"/>

    <!-- Register NCover so it's available for TypeMock. -->
    <Exec Command="regsvr32 /s &quot;Relative\Path\To\NCover\CoverLib.dll&quot;"/>

    <!-- Start TypeMock and link it with NCover. -->
    <TypeMockStart Link="NCover"/>

    <!-- Enumerate the test assemblies you'll be executing with NUnit. -->
    <CreateItem Include="Your\Build\Ouptut\*.Test.dll">
      <Output TaskParameter="Include" ItemName="UnitTestAssemblies"/>
    </CreateItem>

    <!-- Create the folder where unit test and coverage logs will go. -->
    <MakeDir Directories="$(BuildLogDirectory)"/>

    <!-- Run NUnit through NCover so profiling happens. -->
    <!-- Note the use of "batching" so this is equivalent to a "foreach" loop in MSBuild. -->
    <NCover
      ToolPath="Relative\Path\To\NCover\"
      CommandLineExe="Relative\Path\To\NUnit\nunit-console.exe"
      CommandLineArgs="&quot;%(UnitTestAssemblies.FullPath)&quot; /xml=&quot;$(BuildLogDirectory)\%(UnitTestAssemblies.Filename)-results.xml&quot;"
      AssemblyList="MyAssembliesToProfile"
      LogFile="$(BuildLogDirectory)\%(UnitTestAssemblies.Filename)-ncover.log"
      RegisterProfiler="false"
      CoverageFile="$(BuildLogDirectory)\%(UnitTestAssemblies.Filename)-coverage.xml"/>

    <!-- Stop TypeMock and unregister NCover. -->
    <CallTarget Targets="test-finally"/>

    <!-- Get all of the coverage logs and aggregate them with NCoverExplorer. -->
    <CreateItem Include="$(BuildLogDirectory)\*-coverage.xml">
      <Output TaskParameter="Include" ItemName="CoverageReports"/>
    </CreateItem>
    <NCoverExplorer
      ToolPath="Relative\Path\To\NCoverExplorer\"
      ProjectName="YourProjectNameHere"
      ReportType="4"
      Sort="CoveragePercentageAscending"
      Filter="None"
      OutputDir="$(BuildLogDirectory)"
      XmlReportName="CoverageReport.xml"
      HtmlReportName="CoverageReport.html"
      ShowExcluded="True"
      SatisfactoryCoverage="75"
      FailMinimum="True"
      CoverageFiles="@(CoverageReports)"/>

    <!-- In case one of the tests fails, make sure to stop TypeMock and unregister NCover. -->
    <OnError ExecuteTargets="test-finally"/>
  </Target>

  <!-- Stopping TypeMock and unregistering NCover is a separate target because it has to happen -->
  <!-- regardless of success or failure of the unit tests. Like the "finally" in a "try/finally" block. -->
  <Target Name="test-finally">
    <TypeMockStop/>
    <Exec Command="regsvr32 /u /s &quot;Relative\Path\To\NCover\CoverLib.dll&quot;" ContinueOnError="true"/>
  </Target>
</Project>

In the example, notice how the steps for stopping TypeMock and unregistering NCover have been placed in a separate target called “test-finally” since it’s used a lot like a try/finally block. That’s the sort of thing we’re trying to emulate. You’ll also notice that we’re using MSBuild “batching” to run each test assembly through NCover and generate an individual coverage log.

Additional notes:

• Obviously you’re going to need to change the paths and other placeholder parameters to fit your build.
• If you’ve got TypeMock installed on dev machines and on the build box, you can skip the TypeMockRegister task and just start/stop TypeMock.
• If you’ve got NCover installed on dev machines and on the build box, you don’t need to execute “regsvr32” to register/unregister NCover. As long as NCover is registered before you start TypeMock, you’re OK.