Notice! This post is more than a year old. It may be outdated.
Sometimes in unit testing you need to have control of the current time. But when your code that’s under test depends on the current system time, it’s not always that easy to achieve. As developers, we’re taught to favor loosely coupled code. I’m sure you’ve heard of the dependency inversion principle. It states that you shouldn’t depend upon concrete classes but you should depend upon abstractions. Taking that into account, time should also be considered as a dependency.
Current time as a dependency
Think of the following situation. You’re designing a class which needs to check the current time. Let’s say you’re working on a message board (e.g. Discourse) and you would like to congratulate the user on his or her birthday when they log in. The class would need to check the current date and compare it against the user’s entered birthday. If they’re equal, it should display birthday congratulations.
But how would you unit test that class? If your implementation retrieves the current system time, then you have tightly coupled your class to the server’s clock. To avoid that, think of the current time as one of many dependencies your class has. You can use familiar techniques such as dependency injection to pass a replaceable clock to your class. In a production environment, your class can use a clock that retrieves the current system time. In unit tests, you can pass in a clock that reports a given fixed time.
Instances of this class are used to find the current instant, which can be interpreted using the stored time-zone to find the current date and time. As such, a clock can be used instead of
Instead of retrieving the current time from the system, you can ask the time from the clock object.
The primary purpose of this abstraction is to allow alternate clocks to be plugged in as and when required. Applications use an object to obtain the current time rather than a static method. This can simplify testing.
During testing you can use the
Clock.fixed static method to get a clock that always reports a fixed time.
To make things more clear, let’s have a look at a simple example of a class which has a replaceable clock.
ReplaceableClockDemo has an instance field to hold a reference to a
Clock. It can be passed in via the class’ constructor. This allows the client to provide its own clock implementation. In the main method you can see how this class could be used in a production environment. When the class is instantiated, a reference to a system clock is provided.
In unit testing though, you’re free to pass in a fixed clock. As a result, you have full control over the current time in your tests and you don’t have to rely on static methods to set the current system clock.
What about setting the current time via a static method?
If you’ve ever used Joda-Time, then you’re probably familiar with the
DateTimeUtils.setCurrentMillisFixed() static method. It sets the current time but it does this globally. Whenever
currentTimeMillis() is queried, the same fixed millisecond time will be returned.
In theory, this approach works but it has some flaws in my opinion. First of all, you need to make sure you reset the time after each test. Otherwise, some tests that come afterwards might use the same static resource and get a time they did not expect to receive, resulting in test failures. You should not depend on the order of your tests to always be the same. Tests, that are independent are easier to maintain and you also have the added benefit of being able to run them in parallel.
Coupling your code tightly to the system’s clock prevents you from writing reliable unit tests. According to the dependency inversion principle we should depend upon abstractions. Design your classes in a way that you can pass them a desired concrete implementation of time. This allows you to easily set a fixed time in unit tests.