How to Unit Test a C++/WinRT Component? Preferably with Code Coverage - visual-c++

Question

I'm in the process of writing some new C++/WinRT based components in order to replace some much older C++/CX code. The goal is to be able to use third-party C++ tools that don't understand CX (static code analyzers, etc).
However the first step in the journey is to ensure I can properly unit test my own code. Unit testing C++/CX code typically used the "C++ Unit Test App" project type, which is C++/CX based and has its own issues (lack of code coverage support, run all required before tests show up in the explorer, stability, etc)
Browsing through the available project types in Visual Studio 2017, I did not see a unit test project template for C++/WinRT based projects. Is my only option to use the "C++ Unit Test App" template with all its failings, or is there another way to build tests for a C++/WinRT library?
Perhaps there is a way to configure either the "Native Unit Test Project" or "Google Test" project templates to support what I'm looking for?
Ideally what I'm looking for is something that doesn't require launching a UI, is pure C++(/WinRT), and supports Visual Studio's Code Coverage Analysis.

Answer 1

There is no unit test project that is specific to C++/WinRT, much like there isn't one for other libraries like STL. I would recommend Catch2 as it supports C++17 (a requirement for C++/WinRT) and works well on Windows. It is also what we use for testing C++/WinRT itself. Catch2 is nice because it helps you create a simple console app that acts as the test driver that includes all of the tests.
For code coverage I don't have a strong recommendation, but if you are using Visual Studio then you might want to try VSInstr. It can be used for code coverage and produces a report that can be viewed with Visual Studio.
Make sure your code is built using the /profile linker option. This will ensure that profile hooks are included in a dedicated section of the PE file. Next, run vsinstr to instrument any of the binaries you're interested in (that were previously built with /profile):
vsinstr /coverage tests.exe
Now run vsperfcmd to begin collecting coverage data:
vsperfcmd /start:coverage /output:report
Run the code as normal. For Catch2, you can simply run the executable at the command line. Then you need to stop the collection as follows:
vsperfcmd /shutdown
And you're done. You can now view the report in Visual Studio:
devenv report.coverage
Hope that helps. Again, this is not specific to C++/WinRT and since C++/WinRT is a header-only library you are liable to get a lot of noise that is unrelated to your specific project. I haven't found a good way to deal with that yet.

Answer 2

Expanding on my comment to #KennyKerr's answer for those that are interested...
If you are planning on using Catch2 as recommended, then the C++/WinRT Windows Console Application template is a great starting point. Pretty much all you have to do is tweak the main() to setup Catch2 and start writing your test cases. My only complaint is that the C++/WinRT templates don't allow you to add Windows Runtime Component project references via the UI (must be done by editing the vcxproj). There is probably a similar problem adding NuGet package references.
As noted in my comment above, there is a Catch2 test adapter for Visual Studio 2017/2019 in the marketplace. Be aware that it requires a .runsettings file to enable the adapter and to tell it which projects are Catch2 test applications (via a regex). Without a properly configured runsettings, it will not find your tests. I also had to increase the discovery timeout, otherwise it "forgot" my tests occasionally.
With regards the code coverage, when using Visual Studio you can configure the code coverage to include/exclude functions in the .runsettings file. See Microsoft's Site for details. For myself I added the following in the CodeCoverage section and it works pretty well so far:
<Functions>
<Include>
<Function>.*YourNamespaceHere.*</Function>
</Include>
<Exclude>
<Function>winrt.*GetRuntimeClassName</Function>
<Function>winrt::impl.*</Function>
<Function>winrt::(?!YourNamespaceHere).*</Function>
</Exclude>
</Functions>
For those that are trying to test a C++/WinRT Windows Runtime Component like me, and have code that is not exposed as part of the WRC interface, here is what I did to make that testable...
Create a C++ Shared Items Project
Move all of the code for your Windows Runtime Component (WRC) project into the shared items project, and out of the WRC project. Going forward, only add/remove files from the shared project. That way you don't have to touch the WRC or Test projects when files are added/removed.
Add a reference to this shared items project in both your original WRC project, and your test project
Make sure your test project and WRC project are configured similarly with respect compile settings and project/NuGet references
Edit the test project and ensure the RootNamespace is configured the same as the WRC project (probably has to be done via your favorite editor). This is required otherwise the generated headers will be prefixed with the namespace, and thus won't be found by the shared code.
(Optional for Code Coverage) In the test project, enable profiling (Linker > Advanced > Profile > Yes)
You should now be able to write tests that exercise the private code. As to whether or not this is the best approach, I leave to the reader. It works for me, and the code I'm testing is simple enough that I'm not overly concerned with the project definitions not aligning perfectly. Your mileage may vary.
I will note that the above can also be used to make the "Native Unit Test Project" work with C++/WinRT, you just have the extra steps of integrating the C++/WinRT bits into the test project first.