Steve McConnell Text Banner

 

 


Best Practices
IEEE Software, Vol. 13, No. 1, January 1996

Who Cares About Software Construction?


Software practitioners are subjected to a barrage of advice about effective development practices. The search for effective practices--programmers' gold--can be almost as treacherous as the search for the precious yellow metal itself. Some mediocre practices are overhyped and don't pan out; many valuable practices are buried under other practices' hype. This column aims to separate the gold from the ore by providing a practitioner's appraisal of past and present development practices. Future columns will take up practitioner-oriented topics ranging from "Whatever happened to information hiding?" to "The estimation story--Defending unpopular estimates."
-- Steve McConnell

I am going to admit something that people rarely admit in refereed software journals these days:

I like coding.

I admit it, and I'm not embarrassed. I don't hang my head when I say I enjoy low-level design, unit testing, debugging, and optimization. I do not find these low-level construction activities even a little demeaning. I often find them invigorating. Software construction is an important activity that deserves more respect than it has received lately.

At one time, software development and coding were thought to be one and the same. But, as distinct activities in the software-development life-cycle have been identified, the best minds in our field have spent their time analyzing and debating methods of architecture, project management, requirements analysis, design, and quality assurance. The rush to study these newer areas has left code construction as the neglected step-child of software development.

Construction has also been neglected by researchers and writers because of the mistaken idea that, compared to other software development activities, construction is a relatively mechanical process and presents little opportunity for improvement. Nothing could be further from the truth. Construction is not at all mechanical. Sometimes a formal architecture addresses only the top level of system decomposition, and design work is intentionally left for construction. Other times a program's design is supposed to be detailed enough to provide for fairly mechanical coding, but there are always gaps and the coder usually designs part of the program, officially or otherwise.

The reduction in attention to code construction has been exacerbated by the treatment of coding as the dirtiest grunt work of the software-development world. An entry-level programmer in a large organization is typically assigned to code routines that have been specified and designed by someone higher up on the corporate ladder. After a few years, the programmer is promoted to architecture, requirements analysis, or project management and a few years after that may proclaim with pride that it has been years since he or she has written any code.

CENTRAL ROLE. The irony of the shift in focus away from construction is that construction is the only software development activity that's guaranteed to be done. Right or wrong, it's possible to assume requirements rather than analyze them. Architecture can be shortchanged rather than designed. System testing can be abbreviated or skipped rather than carefully planned and executed. But no matter how rushed or poorly planned the project is, construction cannot be dropped. If there's going to be a program, there has to be construction.

Because construction is the only activity that must be done, code is often the only accurate description of the software available, which makes it imperative that the source code be of the highest possible quality. Consistent application of detailed source-code techniques makes the difference between a Rube Goldberg contraption and a polished, correct, and informative program. Such detailed techniques must be applied as the code is initially constructed because it's virtually impossible to go back later and retrofit the thousands of picky details that spell the difference between a maintainable program and a failure.

Even when a program is developed using effective practices, construction typically makes up about 80 percent of the effort on small projects and 50 percent on medium projects. Construction accounts for about 75 percent of the errors on small projects and 50 to 75 percent on medium and large projects. (This data is approximate for detailed design, coding, unit testing, and debugging. See, for example, Programming Productivity by Capers Jones, McGraw-Hill, 1986.) Any activity that accounts for 50 to 75 percent of errors presents a clear opportunity for improvement.

Some commentators have suggested that although construction errors account for a high percentage of total errors, they tend to be less expensive to fix than the errors caused by analysis and design. The implication is that they are therefore less important. The claim that construction errors cost less to fix is true, but it is misleading because the cost of not fixing them can be incredibly expensive. Gerald Weinberg reported ten years ago that the three most expensive programming errors of all time each cost hundreds of millions of dollars and that each was a one-line, coding-level mistake ("Kill that Code!" Infosystems, Aug. 1983). A perusal of any recent "Risks to the Public" section of Software Engineering Notes shows that little has changed since Weinberg made his report. Errors in those single lines might be less expensive to fix than errors in analysis or design, but that obviously does not imply that detecting and correcting them should be a low priority.

OPPORTUNITY KNOCKS.What does all this have to do with best practices?

Construction presents an unusually good opportunity to disseminate information about best practices in software engineering. Millions of programmers are already doing construction and can do it better. People without formal training in computer programming have taught themselves to program in Pascal, Basic, C, and C++. Learning about better software construction allows these people to build on what they already know and gives them a foot in the door to other powerful development practices. Moreover, the number of self-taught programmers has been increasing since the beginning of the PC revolution. The more people who do end-user programming, the more risk there will be of expensive, coding-level mistakes, and the more opportunity there will be to improve the average level of software quality by focusing on improved construction practices.

In 1990, the Computer Science and Technology Board stated that the biggest gains in software-development quality and productivity will come from codifying, unifying and distributing existing knowledge about effective software-development practices ("Scaling Up: A Research Agenda for Software Engineering," Communications of the ACM, March 1990). The board concluded that software-engineering handbooks should play a key role in disseminating that knowledge.

In amorphous areas of software engineering such as requirements analysis and design, there are still vigorous debates about which practices work best. Those debates are not informed by much research. But we as a profession have been doing software construction as long as we have been developing software, and a mountain of research about construction has accumulated over the years. It seems possible to codify the knowledge of what makes for effective construction. In 1993, I took a step toward that end by writing Code Complete, a practitioner-oriented handbook of construction practices (Microsoft Press, 1993). But much more work still needs to be done to identify, validate, and document effective construction practices.

RETRAINING GURUS. It is clear from what I have seen on live projects that young, hot-shot PC developers are making the same mistakes that their elders made and learned to avoid 10 or 20 years ago. It is also clear from much of what has been written that the gray-haired software-engineering experts could learn some things from the young hot-shots (after all, there are reasons why they are hot-shots).

School districts put administrators back into the classrooms occasionally, and service companies put top executives back on customer-service duty. I have a dream in which prominent software-engineering experts sign up for construction roles on projects at modern companies like Microsoft, Borland, and WordPerfect. An unbelievable amount has changed in the way that production software is developed during the last 10 years, and a lot of those differences have to be experienced on a live project to be appreciated. Maybe this dream can never be realized because it calls for too much individual financial sacrifice. Maybe it would seem like too much of a step backward for a software-engineering guru to take a construction job. But the industrywide benefit would be amazing. The insights that some of the wisest and brightest people in our field would gain from being back on the front lines of construction for awhile could invigorate the whole industry.

Journals such as this one can help too. Some professional journals still publish code listings that have such poor layout, variable names, and commenting that they are virtually impossible to read. The code listings contain fundamental programming blunders such as using numeric literals rather than named constants. If these code listings were created in a production environment, the developer who wrote the code would be scheduled for remedial training; companies whose code will be read by only a few dozen developers hold their code to higher standards. Professional journals whose code will be read by thousands should take a leadership role in publishing examples of good construction.

Researchers, consultants, and writers who dismiss construction as a trivial activity need to realize that construction, by definition, occupies the central role in software development. Analysis and design are done before construction so that construction can be done effectively. System testing is done after construction to verify that construction has been done effectively. Construction is at the center of the software-development process. It is where the rubber meets the road, and that makes it a uniquely productive area in which to focus our attentions.

Construction is important and will remain so for the foreseeable future. The difference between good and bad construction can mean the difference between project success and failure. The difference between recognizing the key role that construction plays in software development and ignoring it can mean the difference between moving the industry forward and repeating the same old mistakes yet another time.

Editor: Steve McConnell, Construx Software, 11820 Northup Way #E200, Bellevue, WA 98005.
E-mail: steve.mcconnell@construx.com - WWW: http://www.construx.com/stevemcc/

Email me at stevemcc@construx.com.