Sean Killeen

A mind at work. Microsoft MVP and passionate .NET dev. Trying to be all the human I can be in the time I have. (he/him)

Why I Prefer xUnit over MSTest

3 minute read | Suggest an edit | Issue? Question?

I was recently asked to explain1 my preference for xUnit over something like MSTest. To my surprise, while I felt very strongly – and have for some time – I struggled to clearly articulate my reasons. This is an attempt to get that part right. I sourced some help from folks on Twitter, who I’ll be quoting here as well.

First Thing First: Testing (and its Concepts) Matter Most

In my personal opinion, the time of debating whether or not to have automated tests is long past. There are trade-offs, sure. But particularly in a greenfield project with people who aren’t completely new, testing should be commonplace.

Also, while I’m about to contrast two testing frameworks, I can’t stress enough that developers should be able to use either. Devs who are professionals should understand the common concepts and underpinnings of good tests2 and be able to apply them under any framework, and shouldn’t balk at these things due to being forced into any particular framework.

What I Like About xUnit

  • [Fact] vs. [Theory] attributes. A fact is something that should always be true. A theory is something that, if it’s wrong, could be because you fed it bad data.
  • The simplicity of passing data to tests with [InlineData]. I think this a highly readable way to pass data into a test.
  • Forgetting [Setup] and [Teardown]. My tests flow naturally, just like normal classes and methods should. This prevents me from over-complicating things3.
  • None of that gross [ExpectedException]. In xUnit, I can use Assert.Throws<T>, or with a library like FluentAssertions I can catch exceptions right in my test using an Action:
Action act = () => myClass.DoThingThatBlowsUp();

act.ShouldThrow<Exception>().And.Message.Should().Be("BOOM!");
  • The [Trait] attribute to classify tests. Using [Trait], I can specify any arbitrary metadata about a test I want. Long-running test? Check. Hits a database? I can note that. Deals with a certain component? Yep. BDD-Style English sentence? Yuuup.
  • The ability to run xUnit tests in Visual Studio without an extension. I think the fact that you can run Install-Package xunit.runner.visualstudio and then run your tests right in Visual Studio is awesome. It brings xUnit tests to everyone out of the box.
  • Just a normal class library. No special unit test projects, no run list of unit tests. Just a class library that spits out a DLL. Simple.

What I don’t like about MSTest

In fairness, it’s been a while since I’ve really dug into it. If I’m wrong on any of these points, let me know and I’ll correct ASAP. I did my best to Google whether anything had changed. :smile:

  • It feels like a lot of cruft. To dive into a project using MSTest, it takes a lot longer for me to figure out what tests are doing.
  • It encourages bad habits4. Things like specifying certain pieces of build output for a test that won’t run reliably with other test runners, or creating bloated setup & teardown methods that can cause tests to behave unreliably.
  • [ExpectedException] breaks Arrange/Act/Assert. The assertion I’m making is the first thing in the test, before I’ve arranged or acted on anything.
  • Doesn’t support in-line data or parameterized tests. You have to resort to external files to do data-driven tests.
  • The runner is slow. Similar tests always seem to execute more slowly with the MSTest runner. Apparently I’m not the only one who thinks this way:
  • MSTest doesn’t play well with build servers other than TFS. Per a tweet from Ken Egozi:

Bad Reasons to Justify Using MSTest

  • Microsoft uses it. Pssst, they’re actually mostly using xUnit now.
  • It’s the default and that’s good enough. This is a failure of attitude. Woe to the developer who thinks this way. Continuous improvement, particularly in terms of tool-sets, should be second-nature to a developer.

What do you Think?

Am I too harsh on MSTest? Am I missing something? I’d love to hear from you in the comments.

Happy developing!

Special Thanks…

…are due to Sharon Cichelli, Chris Missal, Ken Egozi, and Brad Wilson for engaging in conversation on this as I prepared the post.5

  1. Some might say “justify.” Semantics. :smile: ↩

  2. I’m thinking along the lines of readability, arrange/act/assert, isolation, etc. ↩

  3. IMO, simplicity and readability are hallmarks of good tests. ↩

  4. My opinion, to be fair. But I think others would agree. ↩

  5. Sorry if I forgot you! Let me know and I’ll fix it, promise. :smile: ↩

Updated:

Leave a comment

You may also enjoy