I started working on improving a codebase recently and one of the best ways to do that is to start writing tests. The site is heavily database driven, so a natural place to start adding tests was in the data layer (including ORM mappings, abstractions, etc.) and in the service layer. The site is powered by SQL Server, so I chose to use SQL Server 2012 LocalDB for testing to minimize friction with the existing code. After writing a load of tests, I wanted to get them running in TeamCity and get coverage numbers going. However, it was not as easy as I expected.

Getting a build agent set up was easy enough, and it took about 30-ish minutes total to start running tests. Introducing coverage didn’t take much longer (TeamCity helpfully includes dotCover for coverage), however, the first coverage-enabled build runs didn’t work. Every build raised the following error from dotCover:

[JetBrains dotCover] Coverage session finished with errors: Can't set event mask (COM error 0x80131377)

My first Google searches led to JetBrains’ bug tracker, where I found this bug open about the same issue. I initially contacted one of the commenters, Will Green (@hotgazpacho on Twitter), who told me that he had given up, as he’d gotten the coverage reports working on his local machine, and didn’t have time to plug through getting it working on the CI server.

A little more Google searching (this time looking for the COM error code) led me to Yi Zhang’s blog on MSDN, where he had a post about HRESULT values returned from .NET/the CLR. Looking up the code, I found that 0x80131377 meant CORPROF_E_INCONSISTENT_FLAGS_WITH_HOST_PROTECTION_SETTING.

The long and the short of that value is that I was trying to run .NET code (ie. the profiler) in the database, but host protection (also known as Code Access Security, or CAS) was not having any of it. A quick MSDN search found that the key was to either mark the assembly memebers as trustworthy (not an option, as I don’t own any of the involved assemblies), or to set the database as trustworthy, which I could do.

As part of the test fixture setup, I drop and recreate the database, so as part of that script, I added the following:

ALTER DATABASE UnitTests SET TRUSTWORTHY ON

I also ensured that the test runner was on the right bitness—TeamCity offers auto, x86, and x64 values. I set mine to x64, as my build agent is a 64-bit machine. Once I finished this setup, I re-ran the build, and got coverage resuts! After setting up my inclusions for assemblies to run coverage against, I was all set to go.

I’m going to comment on the JetBrains’ bug with the above suggestion for resolving the issue for others, but I figured I needed to blog it first, so I can give a little more detail than I can on a bug report.

UPDATE: The bitness setup isn’t necessary—as Will reported in the comments, he only had to run ALTER DATABASE UnitTests SET TRUSTWORTHY ON for success.

UPDATE 2: Comments are gone, but Will reported the following, which appears to work reliably for me as well:

OK, I got it to work! The key is to start the SQLLocalDB instance (via the
SqlLocalDB utility) OUTSIDE the test framework context, then immediately run the
tests that connect to the database, then tear down the DB, again with the
SqlLocalDB utility.