Hi Duy, On Sat, 18 May 2019, Nguyễn Thái Ngọc Duy wrote: > Another round because apparently the test case is not perfect. Test cases are never perfect. But at least the good ones are "actionable", i.e. when they fail, you know there is a regression, it helps you figure out easily why it broke, and it helps you fix the regression. Many, many test cases are bad: they break for all kinds of reasons *except* for a regression. They break because an error message was changed. They break because of timing issues. They break because somebody inserted another test case. They break because a missing prereq caused a previous test case not to run, removing a side effect on which the breaking test case relied. Those bad test cases are very frustrating for people who actually look at them. I am one of those people. And if I did not know delightful test suites that are populated with "good" test cases, I would not point out when a "bad" one is contributed. So: it *is* possible to have a good test suite. Let's make ours better than it is right now. Thanks, Johannes