I have a software development background, yet somehow, testing became one of my favorite topics. I became “test-infected” as the Test-Driven Development folks like to say. In my early coaching experiences I was working with teams on platforms that often didn’t have an XUnit framework. Therefore, I found myself having to defined what a good testing framework would look like so that we could build one for the project! This in turn led me to try to understand a good test. Here are my thoughts on this.
These six qualities of tests describe how to make a test as effective and as useful as possible. The qualities are similar in style to the INVEST (Bill Wake) qualities of user stories or ACID database transactions – but they don’t form a nice acronym.
Decisive: The test contains all the information required to automatically determine success or failure. Manual inspection of test results is not necessary to determine success or failure. The test is expressed in a way that produces a pass/fail answer rather than a numerical or qualitative result. Decisive tests are often expressed as assertions.
Valid: The test produces a result that matches the intention for the work artifact under test. Test failure indicates failure in the artifact under test and test success indicates correct operation or configuration of the artifact under test. Simply put: the result of a test should be true.
Complete: The test contains all the information it needs to run correctly with a given test harness and work artifact under test. The test performs all activities and provides all the data necessary. The test does not require any input outside of itself in order to run.
Repeatable: The test always gives the same results if the test harness and the artifact under test are the same. The test is created in such a way that the result is deterministic. Even if the system under test is not deterministic, the test should be created to account for that (possibly with statistical analysis) and produce a deterministic result.
Isolated: The test result is not affected by other tests run before it nor does a test affect the results of tests run after it. The test and the test harness work together to clean up after every test is run. A collection of tests can be run in any order and always produce the same results. Any test that depends upon the results or side-effects of a previous test is not isolated.
Automated: The test requires only a start signal in order to run to completion in a finite amount of time. No manual intervention is required after the test is started. Tests that are automated can be put together into a test suite and run one after another without human intervention. Automation of tests requires the creation of a test harness system.
Update 20060106 – I’ve added a little bit more detail to the above qualities. I also wanted to say a few words about my experience with testing:
I first started doing test-driven development almost seven years ago after reading the book Refactoring by Martin Fowler. I picked it up in Windsor, Ontario while I was driving to San Francisco for my first contract position as an independent consultant. I had to stay the night there due to the immigration office being closed so I read a lot of the book in my motel room that night. By time I got to California four days later, I was totally convinced that I should be using JUnit and refactoring for everything I did. Fortunately my project was amenable to this approach.
Over the next four months I built a Java JMS layer on top of Tibco Rendezvous. In the process I discovered methods for doing unit tests for asynchronous multi-threaded distributed functionality. And my tests satisfied all the above qualities. I delivered code with 100% test coverage and zero defects discovered until more than two years later when a very rare and obscure deadlock issue was found. Suffice it to say, I was convinced from a quality perspective that Test-Driven Development works.
But there is more. To build tests that follow all the above qualities, you need code that is testable. I have come to believe that testability is the most important architectural attribute of code for most software. The implications of having testable code include code that is easily changed, that is verifiable, and that is easy to understand. Refactoring plays a big role here too.
For getting a basic but very practical sense of test driven development, I strongly recommend “Test-Driven Development: By Example” by Kent Beck.
I have also written a couple of articles recently about quality: Quality is Not Negotiable in which I talk about how to handle defects – by always treating them as top priority work items.
Technical Debt in which I expand on this common analogy between lack of testing and debt to show that technical debt is even worse than financial debt!
[This article was originally published on Agile Advice on 17-May-2005]