Wrong Coverage Analysis on nameof variables

Hi

  • versions used
    Version 8.3 (build 34182)

  • error observed
    Creating consts where its value is the same as the name and use it, but SonarQube Code Coverage reports “0.0% Coverage”

  • steps to reproduce

    • Create a file like “MyConstants.cs”
    • Create a const like so: public const string MyConst = nameof(MyConst);
    • Use this constant in your unit tests
    • make a build and an analysis with SonarQube
    • See that will report it as “not covered”
  • potential workaround
    Instead of public const string MyConst = nameof(MyConst); the Zero Coverage will disappear, when changing it to public const string MyConst = “MyConst”; Technically this is the same, but the content of the variable will not be changed or recognized by the compiler, when you don’t use nameof()
    2020-05-28_06h35_32

hi @DominikTouring

thanks for raising the issue.

  • first, what code coverage tool are you using?
  • second, could you please give us the verbose output of the END step? (please run SonarScanner.MSBuild.exe begin /k:"MyProject" /d:sonar.verbose=true as the BEGIN step, and please attach the output of END step)
  • third, could you check inside the coverage file what the entry for ExceptionConstants.ErrorId shows?

Thanks
Andrei

The covarge tool we are using is the TFS 2018 Build step “Visual Studio Test” - Version 2.0 ( https://docs.microsoft.com/en-us/azure/devops/pipelines/tasks/test/vstest?view=azure-devops ) - Or locally Visual Studio 2019

In coverage file, I can see that this class is never included in the coverage file - Maybe because it’s containing only “non testable” allocation and has been optimized away?

However the behavior changes on SonarQube, when not using “nameof()” - It is still not in the coverage file, but not marked as “uncovered” by SonarQube.

The SonarQube Analysis in Verbose mode is attached
SonarVerbose.zip (63.0 KB)

hi @DominikTouring and please accept my apologies for the late reply, it slipped under my radar.

This is indeed interesting. We only take the coverage by parsing the coverage reports, we don’t treat specific language constructs when parsing the files. It would be interesting to see the two coverage reports, the one with nameof() and the one with string and see the difference.