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。

当你说“我们的代码库不适合简单的测试”时，这是代码气味的第一个迹象。编写单元测试意味着您通常以不同的方式编写代码，以使代码更具可测试性。在我看来，这是一件好事，因为多年来我在编写代码时看到我必须为其编写测试，这迫使我提出更好的设计。

简而言之——是的。它们值得你付出每一分努力……在某种程度上。在一天结束的时候，测试仍然是代码，并且很像典型的代码增长，您的测试最终将需要重构，以便可维护和可持续。有一大堆的陷阱!当涉及到单元测试时，但是没有什么，我的意思是没有什么比丰富的单元测试集更能让开发人员更自信地进行更改了。

I'm working on a project right now.... it's somewhat TDD, and we have the majority of our business rules encapuslated as tests... we have about 500 or so unit tests right now. This past iteration I had to revamp our datasource and how our desktop application interfaces with that datasource. Took me a couple days, the whole time I just kept running unit tests to see what I broke and fixed it. Make a change; Build and run your tests; fix what you broke. Wash, Rinse, Repeat as necessary. What would have traditionally taken days of QA and boat loads of stress was instead a short and enjoyable experience.

提前准备，一点点额外的努力，当你不得不开始摆弄核心特性/功能时，它会给你带来十倍的回报。

我买了这本书——它是xUnit测试知识的圣经——它可能是我书架上被引用最多的书之一，我每天都在查阅它:链接文本

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。

根据我的经验，在复杂的软件环境中，单元测试和集成测试是“必须的”。

为了说服团队中的开发人员编写单元测试，您可能需要考虑在开发环境中集成单元测试回归分析(例如，在您的日常构建过程中)。

一旦开发人员知道如果单元测试失败，他们就不必花那么多时间调试它来发现问题，他们就会更有动力去编写这些测试。

这里有一个工具可以提供这样的功能:

单元测试回归分析工具