简体繁体 English

用Java捕获虚假单元测试

[英]Catch Fake Unit Tests in Java

原文 2018-11-14 18:36:53 8 2 java/ unit-testing/ jacoco

I have tests which do not have asserts in my repo, though jacoco gives good coverage. 我的测试没有回购声明，尽管jacoco提供了很好的覆盖范围。 Is there a way to detect tests like this, other than better code reviews ? 除了更好的代码审查之外，还有没有其他方法可以检测到这种测试？

2 个解决方案

Use PMD. 使用PMD。 It has a standard rule for unitTests without any asserts. 对于没有任何声明的unitTests，它具有标准规则。

If your intent is only to find trivially wrong tests, namely those that lack any assert statement, checkers like PMD might do the job. 如果您的目的只是查找简单的错误测试，即那些缺少任何assert语句的测试，则PMD等检查程序可能会胜任。 You will likely get some false positives, namely for tests where the actual test goal is just to ensure that the SUT does not throw an exception (a case mentioned in the comments by Peter Lawrey). 您可能会得到一些误报，即对于那些实际测试目标只是为了确保SUT不会引发异常的测试（Peter Lawrey的评论中提到的一种情况）。

There are, however, many more problems in tests that are not as simple to find: An assertion could just be wrong ( assertTrue(isPrime(9)) ),or, the assertion may only address a part of the relevant aspects (like, when dealing with rational numbers, only checking the numerator but not the denominator). 但是，在测试中还有许多问题并不容易找到：一个断言可能是错误的（ assertTrue(isPrime(9)) ），或者，该断言可能只解决了相关方面的一部分（例如，处理有理数时，仅检查分子，而不检查分母。

For the detection of such quality problems there also exist approaches, like mutation testing, which can help to some extent. 为了检测这种质量问题，还存在一些方法，例如变异测试，可以在一定程度上有所帮助。 And, while writing the tests in the first place: Using a test first development process ensures that a test at least failed once. 而且，在首先编写测试时：使用测试优先开发过程可确保测试至少失败一次。

However, a test suite may have additional problems that are not related to the set of tests that are executed, but to non-functional quality criteria: Test execution time, maintainability of the test suite, expressiveness of the diagnostic output in case of a test failure etc. 但是，测试套件可能存在其他问题，这些问题与已执行的测试集无关，但与非功能质量标准有关：测试执行时间，测试套件的可维护性，测试情况下诊断输出的表达能力失败等

When you are interested in detecting these kinds of problems, I would argue reviews are unavoidable. 当您有兴趣检测这类问题时，我认为评论是不可避免的。