Large test suites often contain many tests that are absolutely redundant. They overlap with other tests so much that they only find bugs that these other tests already find. Such redundant tests are not only useless, they are harmful: They slow down test execution without adding value in terms of bug-finding capability. And even worse, they eat up effort to maintain that would be better spent elsewhere. So let’s remove them!

But how can we detect redundant tests in our large test suites? And how can we be sure that they really provide no value?

In my talk, I will present several approaches to identify redundant tests. I will share their research background, our empirical studies and our experiences with how well they work under which conditions.

Value for the audience:
From the discussions I had with participants at SWQD in the last years, I know that many are struggling with rising maintenance costs of automated test suites: It costs a lot of effort to inspect flaky tests, debug failures caused by application changes (instead of revealed bugs) to keep the tests alive. This effort cannot be spent on new tests that often have a much higher chance of finding critical new bugs.

While many are aware, in an abstract way, that they are facing this problem, it is still hard to know which tests exactly can safely be removed to save costs. This leads to inaction, especially because incentives are often as such: Nobody gives you much praise, if you save on maintenance effort. But you will certainly receive a lot of criticism, if you delete tests that then allow bugs to slip through!

I hope that my talk frames the problem and the available solutions in an actionable way that helps to break this stalemate. I am optimistic that the research that we did to empirically validate (and invalidate) the approaches helps to increase the trust in test minimization approaches. I have been following a similar strategy (get inspired on new topics by talking to participants at SWQD, then supervising research projects on them, then building a talk on the results) in the past at SWQD (on different topics).