Initial Results on Counting Test Orders for Order-Dependent Flaky Tests Using Alloy

Flaky tests can seemingly nondeterministically pass or fail for the same code under test. Flaky tests are detrimental to regression testing because tests that pass before code changes and fail after code changes do not reliably indicate problems in code changes. An important category of flaky tests is order-dependent tests that pass or fail based on the order of tests in the test suite. Prior work has considered the problem of counting test orders that pass or fail, given relationships of tests within a test suite. However, prior work has not addressed the most general case of these relationships. This paper shows how to encode the problem of counting test orders in the Alloy modeling language and how to use propositional model counters to obtain the count for test orders. We illustrate that Alloy makes it easy to handle even the most general case. The results show that this problem produces challenging propositional formulas for the state-of-the-art model counters.