Gerard Meszaros is an Agile software development consultant, trainer, author and occasional conference speaker. His book, xUnit Test Patterns: Refactoring Test Code is a recent Jolt award winner and is becoming a standard reference on the subject of unit testing. Gerard will be speaking at the Agile 2009 conference with a session on "From Concept to Backlog," covering the fuzzy front end of software projects. We interviewed Gerard just before the conference on how the book developed, where the patterns can be applied, and his view of the future of testing and Agile.
Matt Heusser: Obviously you have a background in programming and testing. Tell us a little more about where you've been, and how that brought you to write xUnit Test Patterns.
Gerard Meszaros: My background is as a developer, project manager, and software architect building large telecom systems. I was involved in reviewing the original Design Patterns (Gamma et al) and the first patterns conference (PLoP) in the mid 1990s. I also lived through the transition from ad hoc object-oriented methods to UML and RUP, and it seemed like it was getting harder and harder to develop all the artifacts (and teach others how to do it).
I knew Kent Beck from the patterns community and laughed when he first said he was doing “extreme programming.” As an architect, it seemed completely anathema to everything I valued. But in deeper discussions with Kent, I saw that our values were the same. This led me to investigate further how to apply the ideas.
Matt: Why test patterns?
Gerard: There is plenty of literature on design patterns for production code. As I wrote in the preface of my book, I started doing extreme programming (XP) and test-driven development (TDD) in the late 90s; I found that how we wrote the tests had a huge bearing on how much they helped (or hindered) rapid development.
We were really on the bleeding edge in those days so there weren’t a lot of resources available to help us. We compared notes at the early Agile/XP conferences and found that a lot of us were landing on more or less the same techniques to address the issues.
xUnit Test Patterns is the book I wish I had when I was starting out. I’ve been very gratified by other practitioners saying exactly the same thing when they reviewed the book.
Matt: How is unit testing different than traditional, black-box testing? Do you think developers could stand to benefit from the traditional black-box test list (such as bounds and equivalence classes)?
Gerard: In some ways, it is very similar and in others it is quite different. We use a lot of the standard test techniques to identify the tests that we need. But we write the tests for a very different purpose.
I think of the automated unit tests (and automated story tests) as “bug repellent” rather than “bug detection.” Traditional testing is about finding bugs while automated unit testing, done in conjunction with Test-Driven Development, is all about preventing the bugs from happening.
Matt: Likewise, are xUnit test patterns also transportable to business-facing level of testing? What about load, performance, and exploratory testing?
Gerard: Yes, testers like Lisa Crispin reviewed my book and told me that they found it useful when designing their business-facing tests.
Regarding the other kinds of tests: A lot of the techniques I describe can be applied to para-functional tests such as load and performance testing. Exploratory testing by definition cannot be fully automated but testers doing exploratory testing can use automated test tools or scripts to help them execute the tests they define on the fly.
Matt: Where do you see limitations of the test patterns? Which code do you refuse to test? That is, do you see unit tests having a place for developers using more bare-metal tools like shell-scripts, or assembler code?
Gerard: All techniques have limitations, especially when they are abused or overused. The biggest hazard with automated unit testing is writing tests that are hard to understand and which are therefore hard to maintain as you evolve the code. A lot of the techniques I describe are ways to avoid this issue.
As for using TDD and automated unit tests in bare-metal tools like C, the higher the overhead of debugging, the more benefit there is to doing TDD because preventing mistakes (also known as bugs) saves even more effort. One area particularly ripe for the application of TDD is embedded software. Developers can save a lot of effort by removing most of the bugs while working in an IDE so that when they deploy the code into the hardware the only bugs left are ones related to integration, not basic logic errors. This requires building some test infrastructure and layering in the software properly to allow stubbing out of the actual hardware interfaces -- but the payback is huge.
Matt: Does unit testing change when you shift paradigms? For example, should C code be unit tested differently than Java? And if so, how?
Gerard: Yes, C should be tested even more thoroughly than Java or C# because the compiler won’t catch as many mistakes! Likewise for dynamic scripting languages like Ruby or Perl; it is easy to introduce bugs that will only be found at runtime, so running all the code regularly via automated tests is crucial.
Matt: What measures and metrics do you think are appropriate for developer tests?
Gerard: When doing TDD, statement coverage of the code being tested should be very high, ideally 100%. I’m not a metrics guy but I would suspect that test code should have a very low Cyclomatic Complexity metric because it should contain very little conditional logic. To me, the ultimate measure of the quality of the test code is how easy it is to read, because it will be read a lot!
Matt: In the past, we've heard a lot of confusing terms. Unit tests, acceptance, system, integration ... how do you slice and dice your definitions of tests?
Gerard: Terminology is always an issue. There is no single definition we can all agree on for many of these terms. Trying to standardize the terminology is one of the reasons why I wrote the book, but not every agrees on the need to have agreement on crisp definitions. Personally, I subdivide tests based on the kinds of characteristics we are verifying (functionality, para-functional qualities) and the granularity of the system under test. So I have functional unit tests, component tests, system tests, and workflow/integration tests. Acceptance tests and regression tests are just two particular uses of any one tests at different points in the lifecycle of the code.
Matt: What attracts you to Agile 2009 over other conferences?
Gerard: I get to get back together with many of the other thought leaders in the Agile space and we can compare notes on what we have learned in the past year. Most everyone who’s anyone is there.
Matt: And what will you be speaking about?
Gerard: I’m giving a tutorial on xUnit Test Patterns (of course!) as well as one called From Concept to Backlog. The latter describes the overall process between when a business person gets an idea for how to make or save money with software and when they show up in the team room with a stack of story cards and a wad of cash (the budget). Most Agile methods ignore this part of the project, so there is a lot of confusion and mythology about what should or should not happen in this very important time period. As a result, a lot of projects involving people new to Agile really flounder in the early stages. This can be prevented by some appropriate planning and activities. And no, planning is not banned by Agile!
Matt: How do you see the future of testing tools and test automation frameworks such as FitNesse and Concordion?
Gerard: I think we are in the middle of a major growth spurt for these tools and new competitors like Cucumber. Storytest-Driven Development (or Acceptance Test-Driven or Example-Driven) is a key success factor in Agile projects. We are finally getting some tools that provide decent support for the process.
Of course, there is still a lot of room for improvement, and the various tools are pushing each other to get better -- which is great!
Matt: What do you think are the next steps that the Agile movement will take in the next few years?
Gerard: If I were good at predicting the future I’d be rich playing the stock market or the horses -- or would be a hugely successful waterfall project manager! Since I’m neither of these, all I’m prepared to say is that I suspect we’ll see continued refinement of our understanding of what makes Agile software development effective.
Part of this is the mixing in of lean concepts such as Pull, Flow, and Waste Reduction. Personally, I have been pushing hard the incorporation of UxD (User/Usage centered Design) practices into the Agile process; I think we’ll see a lot more of this. I see that one of the conference keynotes is by a UxD person (Jared Spool), which is great.