作为一名物理专业的学生，我非常有动力去证明我的代码是可以正常工作的。您可以从逻辑上证明这一点，随着实现变得更加复杂，难度会急剧增加，或者您可以通过良好的测试对功能进行(尽可能接近的)经验证明。

如果不提供函数的逻辑证明，就必须进行测试。唯一的选择是说“我认为代码可以工作....”

[我有一个观点，我不能在上面看到]

“每个人都在进行单元测试，他们不一定意识到这一点——事实”

想想看，你写了一个函数来解析一个字符串并删除新的行字符。作为一个开发新手，你要么在命令行中通过Main()实现它来运行几个用例，要么用一个按钮组合一个可视化前端，将你的函数绑定到几个文本框和一个按钮上，然后看看会发生什么。

这就是单元测试——基本的和糟糕的组合在一起，但是你测试了一些情况下的代码段。

你写一些更复杂的东西。当您抛出一些用例(单元测试)并将其调试到代码中并进行跟踪时，它会抛出错误。你在浏览过程中查看这些值，并决定它们是对还是错。在某种程度上，这是单元测试。

这里的单元测试实际上是采用这种行为，将其形式化为结构化模式并保存，以便您可以轻松地重新运行这些测试。如果您编写了一个“适当的”单元测试用例而不是手动测试，那么它所花费的时间是相同的，或者随着您的经验的增加可能会更少，并且您可以一次又一次地重复它

单元测试绝对是值得付出努力的。不幸的是，您选择了一个困难的(但不幸的是常见的)场景来实现它。

从单元测试中获得的最大好处是从头开始使用它——在一些精选的小项目中，我有幸在实现类之前编写单元测试(此时接口已经完成)。通过适当的单元测试，您可以在类还处于婴儿期时就发现并修复它们，而不是在将来毫无疑问会集成到复杂系统中的任何地方。

如果您的软件是完全面向对象的，那么您应该能够在类级别上添加单元测试，而不需要太多的努力。如果您没有那么幸运，您仍然应该尽可能地尝试合并单元测试。确保当你添加新功能时，新部分都有清晰的接口，你会发现单元测试让你的工作变得更容易。

我同意与大多数人相反的观点: 不编写单元测试是可以的 特别是重原型的编程(例如AI)很难与单元测试相结合。

thetalkingwalnut问道: 有什么好方法可以让团队中持怀疑态度的开发人员相信单元测试的价值?

Everyone here is going to pile on lots of reasons out of the blue why unit testing is good. However, I find that often the best way to convince someone of something is to listen to their argument and address it point by point. If you listen and help them verbalize their concerns, you can address each one and perhaps convert them to your point of view (or at the very least, leave them without a leg to stand on). Who knows? Perhaps they will convince you why unit tests aren't appropriate for your situation. Not likely, but possible. Perhaps if you post their arguments against unit tests, we can help identify the counterarguments.

It's important to listen to and understand both sides of the argument. If you try to adopt unit tests too zealously without regard to people's concerns, you'll end up with a religious war (and probably really worthless unit tests). If you adopt it slowly and start by applying it where you will see the most benefit for the least cost, you might be able to demonstrate the value of unit tests and have a better chance of convincing people. I realize this isn't as easy as it sounds - it usually requires some time and careful metrics to craft a convincing argument.

单元测试是一种工具，就像任何其他工具一样，应该以这样一种方式进行应用，即收益(捕捉错误)大于成本(编写它们的工作)。如果它们没有意义，就不要使用它们，记住它们只是你工具库的一部分(例如检查、断言、代码分析器、形式化方法等)。我告诉开发者的是:

They can skip writing a test for a method if they have a good argument why it isn't necessary (e.g. too simple to be worth it or too difficult to be worth it) and how the method will be otherwise verified (e.g. inspection, assertions, formal methods, interactive/integration tests). They need to consider that some verifications like inspections and formal proofs are done at a point in time and then need to be repeated every time the production code changes, whereas unit tests and assertions can be used as regression tests (written once and executed repeatedly thereafter). Sometimes I agree with them, but more often I will debate about whether a method is really too simple or too difficult to unit test. If a developer argues that a method seems too simple to fail, isn't it worth taking the 60 seconds necessary to write up a simple 5-line unit test for it? These 5 lines of code will run every night (you do nightly builds, right?) for the next year or more and will be worth the effort if even just once it happens to catch a problem that may have taken 15 minutes or longer to identify and debug. Besides, writing the easy unit tests drives up the count of unit tests, which makes the developer look good. On the other hand, if a developer argues that a method seems too difficult to unit test (not worth the significant effort required), perhaps that is a good indication that the method needs to be divided up or refactored to test the easy parts. Usually, these are methods that rely on unusual resources like singletons, the current time, or external resources like a database result set. These methods usually need to be refactored into a method that gets the resource (e.g. calls getTime()) and a method that takes the resource as a argument (e.g. takes the timestamp as a parameter). I let them skip testing the method that retrieves the resource and they instead write a unit test for the method that now takes the resource as a argument. Usually, this makes writing the unit test much simpler and therefore worthwhile to write. The developer needs to draw a "line in the sand" in how comprehensive their unit tests should be. Later in development, whenever we find a bug, they should determine if more comprehensive unit tests would have caught the problem. If so and if such bugs crop up repeatedly, they need to move the "line" toward writing more comprehensive unit tests in the future (starting with adding or expanding the unit test for the current bug). They need to find the right balance.

重要的是要认识到单元测试并不是万能的，而且存在太多单元测试这样的事情。在我的工作场所，每当我们做一个经验教训，我不可避免地听到“我们需要写更多的单元测试”。管理层点头表示同意，因为“单元测试”==“好”这句话已经被灌输到他们的头脑中了。

However, we need to understand the impact of "more unit tests". A developer can only write ~N lines of code a week and you need to figure out what percentage of that code should be unit test code vs production code. A lax workplace might have 10% of the code as unit tests and 90% of the code as production code, resulting in product with a lot of (albeit very buggy) features (think MS Word). On the other hand, a strict shop with 90% unit tests and 10% production code will have a rock solid product with very few features (think "vi"). You may never hear reports about the latter product crashing, but that likely has as much to do with the product not selling very well as much as it has to do with the quality of the code.

Worse yet, perhaps the only certainty in software development is that "change is inevitable". Assume the strict shop (90% unit tests/10% production code) creates a product that has exactly 2 features (assuming 5% of production code == 1 feature). If the customer comes along and changes 1 of the features, then that change trashes 50% of the code (45% of unit tests and 5% of the production code). The lax shop (10% unit tests/90% production code) has a product with 18 features, none of which work very well. Their customer completely revamps the requirements for 4 of their features. Even though the change is 4 times as large, only half as much of the code base gets trashed (~25% = ~4.4% unit tests + 20% of production code).

我的观点是你必须传达你理解单元测试太少和太多之间的平衡——本质上你已经考虑了问题的两面。如果你能说服你的同事和/或你的管理层，你就获得了信誉，也许就有更好的机会赢得他们的信任。

还没有人提到的一件事是，让所有开发人员承诺实际运行和更新任何现有的自动化测试。当您重新进行自动化测试并发现由于新的开发而出现故障时，会失去很多价值并使自动化测试非常痛苦。因为开发人员已经手动测试了代码，所以大多数测试不会指出错误，所以花在更新它们上的时间只是浪费。

说服怀疑者不要破坏其他人在单元测试上所做的工作，对于从测试中获得价值更为重要，而且可能更容易。

每次从存储库更新时，花费数小时更新由于新特性而损坏的测试既没有效率也没有乐趣。