如何让 SonarQube 分析单元测试中的代码气味，但不将它们计入覆盖率报告？

Question

I have C++ project being analysed with the commercial SonarQube plugin.

My project has, in my mind, an artificially high code coverage percentage reported as both the "production" source code lines and Unit Test code lines are counted. It is quite difficult to write many unit test code lines that are not run as a part of the unit testing, so they give an immediate boost to the coverage reports.

Is it possible to have the Unit Test code still analysed for Code Smells but not have it count towards the test coverage metric?

I have tried setting the sonar.tests=./Tests parameter (where ./Tests is the directory with my test code. This seems to exclude the test code from all analysis, leaving smells undetected. I would rather check that the test code is of good quality than hope it is obeying the rules applied to the project.

I tried adding the sonar.test.inclusions=./Tests/* in combination with the above. However, I either got the file path syntax incorrect or setting this variable causes a complete omission of the Test code, so that it no longer appears under the 'Code' tab at all as well as being excluded.

The documentation on Narrowing the Focus of what is analysed is not all the clear on what the expected behaviour is, at least to me. Any help would be greatly appreciated as going through every permutation will be quite confusing.

Perhaps I should just accept the idea that with ~300 lines of "production" code and 900 lines of stubs, mocks and unit tests a value of 75% test coverage could mean running 0 lines of "production" code. I checked and currently, my very simple application is at about that ratio of test code to "production" code. I'd expect the ratio to move more towards 50:50 over time but it might not do.

Answer 1

One solution I found was to have two separate SonarQube projects for a single repository. The first you setup in the normal way, with the test code excluded via sonar.tests=./Tests . The second you make a -test repository where you exclude all your production code.

This adds some admin and setup but guarantees that coverage for the normal project is a percentage of only the production code and that you have SonarQube Analysis performed on all your test code (which can also have coverage tracked and would be expected to be very high).

I struggle to remember where I found this suggestion a long time ago. Possibly somewhere in the SonarQube Community Forum , which is worth a look if you are stuck on something.