The promise of testing

When I first learned about automated tests for software, it felt as if a door to a new world had opened up. Automated tests offered the promise of scientific precision and engineering rigor in software development, a process which I thought was limited by its nature to guesswork and trial and error.

This initial euphoria lasted less than a year. The practical experience of creating and maintaining tests for real-world applications gave me many reasons to doubt the promise of automated tests. Tests took a lot of effort to write and update. They often failed even though the code worked. Or the tests passed even when the code did not work. In either scenario, much effort was devoted to testing, without much benefit.

As the number of tests grew, they took longer and longer to run. To make them run faster, more effort had to be devoted to optimizing their performance or developing fancy ways of running them (on multiple cores or in the cloud, for example).

But even with many, many tests and more lines of test code than actual application code, many important features had no tests. This was rarely due to negligence but due to the difficulty of testing many aspects of real-world software.

Finally, bugs still popped up and tests were often written as part of the bug fix to ensure these bugs would not happen again. Just as generals always prepare for the last war, tests were ensuring the last bug didn't happen without helping prevent the next bug.

After several years of facing these challenges, and addressing them with various strategies, I realized that, for most developers, automated tests had become a dogma, and tests were primarily written for their own sake.

To benefit from automated tests, I believe one must consider the cost of testing. In other words, the effort of writing the test must be worth the benefits it offers. What I have learned is that the benefit of tests is rarely to directly prevent bugs, but rather to contribute to improved code quality and organization, which, in turn, will lead to more reliable software. Put another way, although automated tests are closely tied to quality assurance, their focus should be on quality, not assurance. This is just common sense if you think about it. How can we give assurance with automated (or manual) tests that a real-world piece of software, composed of thousands of lines of code, will not have bugs? How can we predict every possible use case, and how every line of code will behave?

Another issue is how to write tests. A number of challenges arise when testing complex applications in the real world. Should you use fixtures or mocks to test models? How should you deal with rack middleware in controller tests? How should you test code that interacts with external APIs and services? This book offers the essentials required to solve problems like these with RSpec, the popular Ruby testing library.

The goal of this book is to help you effectively leverage RSpec's many features to test and improve your code. Although we will limit ourselves to the most pertinent options, I encourage you to consult the official RSpec documentation (http://rspec.info/documentation/) to learn more about all the possible options. You should find it easy to build upon the examples here to develop a custom solution that exactly meets your own needs and preferences.