Practices Supporting Software Craftsmanship
The question then becomes: How can we constantly increase the likelihood that our software is correct from the start? First, let computers do what computers do best, the rote mechanical activities and detailed analyses that are time-consuming and error-prone for humans. The ways we can leverage automated code quality tools leads to a taxonomy of software hygiene.4
4. Thanks to Zee Spencer for his suggestions in this area.
• Style: Whitespace, curly braces, indentation, etc.
• Syntax: Compilation, static analysis
• Simplicity: Cyclomatic complexity, coupling, YAGNI5
5. “You aren’t going to need it” from extreme programming. [XPI, p. 190]
• Solution: Runtime, correctness, whether it works, TDD, BDD
• Scaling the effort: Continuous integration, continuous delivery
Starting with the integrated development environment (IDE),6 we have a wealth of features at our disposal covering Style to Solution. Code completion, syntax highlighting, and project and file organization and management only start the list. Refactoring support and other tool integrations sweeten the pot. If you can, configure all of your tools to give immediate, real-time feedback within your IDE. The less time that passes between when you write the code and when you get the feedback, the closer you come to first-time quality.
6. Or your editor of choice. A number of text editors, such as Sublime Text (www.sublimetext.com), support many IDE-style features in a lighter-weight package.
Do not concern yourself with the placement of curly braces and the amount and type of whitespace. Configure your IDE or your build to do all of your formatting for you. Ideally, your IDE will do it so that it is applied instantly. The next best thing gives that responsibility to a build target that you can run locally at any time.
Further protect yourself with static analysis tools to help with Syntax and Simplicity. Tools like Checkstyle for Java, jshint or jslint for JavaScript, perlcritic for Perl, lint for C and C++, or the equivalent for your language of choice help to check for stylistic problems. More sophisticated tools like PMD7 or FindBugs for Java go beyond formatting and style and explicitly look for everything from unused and uninitialized variables to framework-specific conventions and complexity metrics. Some are even extensible. PMD in particular has a very flexible capability to define rules based on XPath expressions against the abstract syntax tree. PMD also has a module called CPD, the copy–paste detector, that highlights code that was copied when perhaps it should have been refactored to reuse.
7. PMD is a static analysis tool for Java and other languages. It can be found at http://pmd.sourceforge.net. The developers claim that it is not an acronym, but a “backronym.” See http://pmd.sourceforge.net/meaning.html for explanation and possible interpretations.
Code coverage tools help you guide your testing by numerically and graphically showing you what parts of your code have been executed, guiding you toward the right Solution. We will look more closely at code coverage ahead.
Code generation tools will create thousands of lines of code based either on your source or their own domain-specific language, potentially helping with Simplicity and Solution. Tools exist to generate network service interfaces, database access code, and more, even though in many cases the use of code generation is a code smell8 of a missed opportunity for code reuse, also Simplicity. In some cases, you even have tools to help generate your tests. If you trust your code generator, you can skip testing the generated code, saving time and effort on two fronts.
8. Another of Kent Beck’s contributions to software craftsmanship, meaning a symptom of deeper problems.
Finally for tools, continuous integration can run all of these tools along with the build and the tests to provide qualification of your code independent of the local environment, helping with Scaling. The results of the continuous integration builds should be available via the web, on monitors, and through e-mail. Many groups have tied their failing continuous integration results to sirens or flashers to make it obvious. Continuous integration builds should provide results in minutes for best effect. Continuous integration provides yet another form of rapid feedback, trading a little bit of responsiveness for a more comprehensive evaluation.