Pages: pp. 8-11
In his May/June From the Editor column, Steve McConnell asked, How good is software construction? In a word, it stinks. You would think it would be better given that so much has been researched and written. Unfortunately, most people haven't progressed past their first year of work experience. They don't read the literature. The same mistakes are being made in IT shops all over the place.
A recent boss of mine said he hadn't opened an IT book since he left school 25 years ago. It showed. I moved on to a huge project that didn't track schedules against deadlines nor adjust the project plan, and they didn't review code. They were hugely over budget, and the quality was terrible.
After an impressive interview process for my current job, I was looking forward to it—until I got here. The code is terrible. They're trying to test it into shape. The data schema doesn't support the client's needs, and they just try to work around it. The system is fragile, and things break every day. No one manages the team's work or ensures product quality.
So, the state of IT is terrible, yes? Well, yes and no.
Professionally, the industry is terrible. Personally, it's great. The fact that no one else is paying attention to the literature makes it easy to look good. Please don't try to improve things.
James Cherryh, Software developer
How good is software construction? We spend a lot of industry air time (articles, books, conferences, consulting) on human-oriented soft things (requirements, use cases, design, arguments between CMM and the many agile methods) in the life cycle.
But we almost totally ignore the contribution of automated tools and processes to software productivity. If we had to do manually what tools (compilers, configuration management systems) and automated processes (scripts, software builds) do, that contribution would be more visible.
If the use of automation is considered a winning technique, why don't we cut some of industry's losses on the soft side and invest that effort in building more on the automated side? Businesses cut loser projects and fund winner projects. Sports teams cut losers and keep winners. People cut their losses and play to their personal strengths in their careers. Are we doing the same in the software industry, and playing to our strengths?
It seems not. We should spend more time on new, winning automated processes and less time patching up losing human processes.
Kevin Jameson, CEO, Codefast.com; email@example.com
Steve McConnell's May/June column caught my attention—in particular, the question, How do you manage multiple releases? This has been the focus of my technical passion for the past 12 years.
I see fewer and fewer companies who build "just one." I attribute this trend to customers who won't settle for a generic solution that almost solves their problem, embedded systems that have to run on multiple variants of hardware, and increasing demands for integration of "everything to everything." Even though most organizations create and support multiple flavors of their software, the industry still focuses on using technology and techniques for "one-of-a-kind" software engineering.
In the same way that configuration management provides abstractions and tools for managing software "variation in time," we need abstractions and tools to manage software "variation in space"—that is, multiple system variations in a domain space. It is possible to divide and conquer the complex variation management problem (combining configuration management and customization management) into nine simple problems. Some of these problems are addressed by off-the-shelf CM technology, but others represent a void in current practices.
Think of this problem as software mass customization (the sidebar offers an annotated bibliography). Toyota can produce thousands of different Camrys from a single product facility. Their engineers don't focus on the multiple Camry instances but on the single production line that produces them. In the same way, software organizations that build and maintain lots of product variations must shift their focus from individual products to the unified line that produces the different variations of their products. In the same way that manufacturing underwent a revolution from mass production to mass customization, software organizations that effectively embrace software mass customization will gain huge strategic and tactical advantage, thereby driving the whole industry in that direction. In Software Product Lines: Practices and Patterns, Paul Clements and Linda Northrop bring this point home in their wonderful sidebar entitled "E Pluribus Unum"—out of many, one.
Charles Krueger, CEO, BigLever Software; firstname.lastname@example.org
Steve McConnell asks, Why are we so gullible about silver bullets? Fredrick P. Brooks provides the answer in The Mythical Man-Month (Addison-Wesley, 1975): "All programmers are optimists…. 'This time it will surely run,' or 'I just found the last bug.'" We can extend this to "This silver bullet will work." It takes a lot of hard work to become cynical enough to overcome our natural optimism.
Brian Farquhar, Sydney, Australia; email@example.com
I was disappointed to read the July/August article "Software Engineering Is Not Enough." The authors discount the notion of program "simplicity, clarity, and elegance." KISS is one of the most important aspects of good coding practice. Yes, there is nothing simple about a compiler, but the parts that make the compiler must be simple, clear, and elegant. Perhaps they forget that divide and conquer (one of the most important principles in all of computer science, in my opinion) can often result in the most simple, clear, and elegant solutions.
They also claim that "there is rarely such a thing as too much documentation." I have to disagree again. A well-written, clear, and elegant piece of code is much easier to understand than having to wade through thousands of pages of documentation. This is not to say that documentation is not needed, but, like anything else, let's have some moderation.
Fred Rahmanian, Senior consultant, Source Tech Corp.; firstname.lastname@example.org
I do not necessarily disagree with the general point that James Whittaker and Steven Atkin try to make in their July/August article "Software Engineering Is Not Enough." However, I find the example they cite as the result of a "maintenance nightmare" to be singularly inappropriate.
While this "DCB" typedef might appear on the surface to be a hodge-podge of unrelated attributes, in fact it is not. The trained observer will recognize that virtually all the fields are necessary mode settings to be made when establishing an RS232 serial communications link with an arbitrary external device. About two-thirds of them are necessary hardware settings for modern serial communications chips. Most of the remaining third pertain to standard flow control issues in a device driver for a high-speed serial port.
Rather than serve as an example of where frenzied maintenance developers forget design principles, I submit that a typedef for controlling modern serial communications developed from scratch using the best design principles would closely resemble the one shown. If you think the complexity of this domain is artificially inflated due to the example's coming from the Windows API, try typing " stty -all" at your Unix console.
The authors' original argument might in fact apply to the history and evolution of RS232 serial communications itself (with the practically inexcusable variety of formats, handshaking options, error-checking techniques, and so on). But in any case, they're wrong to blame the lowly software maintenance developer.
James J. Besemer, President and chief engineer, Cascade Systems Solutions; email@example.com
James A. Whittaker and Steven Atkin respond:
It was our intention in writing this article to stir debate and discussion; we are pleased to see that we succeeded. It was also our intention to elevate the importance of developers and development skill, which we believe is the main cause for high-quality software and often the least appreciated task in software engineering. Whether one agrees with our specific examples is less important than the fact that discussion is taking place. We firmly stand behind our belief that software development is fundamentally a technical problem and that software processes must center around technology and discipline.
In response to the January/February 2002 From the Editor column "Closing the Gap," I find that we use improved software processes as a defense mechanism in situations where we have to answer "no" to the following, deliberately loosely formulated, questions:
When I was working in a job where I could not say "yes" to these, I was aggressively trying to promote and introduce better software practices. Now, I have changed companies and can say "yes" to all the above. So, the appeal of (new) software engineering methods has dramatically faded.
I would even say that given the good situation I find myself in, any introduction of new process would have to be heavy on the side of "what's in it for me," or I just won't be interested.
But isn't the introduction of new practices a management decision, you ask? No. If I'm the guy who does the designing, testing, or implementing, then anything I don't personally and full-heartedly buy into won't make it. After all, I'm a software developer—I am prepared to quit (again) if my work conditions become stressful or unacceptable. Besides, it's rarely management who comes and says we should spend more time on, say, the design process instead of coding.
I still have a positive approach: I am keen to embrace new practices as long as I'm convinced they're any good.
Stephen Muires, Software engineer, Terma A/S; firstname.lastname@example.org
Send letters to email@example.com. Include your full name, title, affiliation, and email address. Letters are edited for clarity and space.
Software mass customization is a convergence of the software generator and software product line fields. It involves automatically generating individual products in a software product line from abstract product descriptions.
www.biglever.com. The BigLever Software site contains a description of software mass customization, links to related articles, books, and other sites, and an example of commercially available products and services.
C. Krueger, "Easing the Transition to Software Mass Customization," Proc. 4th Int'l Workshop on Software Product Family Eng., Springer-Verlag, 2001, pp. 282-293. An overview of software mass customization, how it relates to conventional software engineering, and efficient techniques for transitioning to this approach.
P. Clements and L. Northrop, Salion, Inc.: A Software Product Line Case Study, Software Engineering Institute tech. report, Carnegie Mellon Univ., 2002. This company adopted software mass customization to manage its product line.
K. Czarnecki and U. Eisenecker, Generative Programming, Addison-Wesley, 2000. A comprehensive book on the issues surrounding software generators. It surveys existing work but also distills information into concise and usable descriptions.
www.semanticdesigns.com. The Semantic Designs site contains examples of commercially available products and services for creating software generators.
J.C. Cleaveland, Program Generators with XML and Java, Prentice Hall, 2001. A guide for creating software generators using some of today's popular software development technology.
www.program-transformation.org/gp. Generative Programming Wiki is an information clearinghouse that the generative programming community maintains in an editable Wiki format.
IEEE Software's "Initiating Software Product Lines" special issue, vol. 19, no. 4, July/August 2002. A good introduction to the field.
www.sei.cmu.edu/activities/plp/plp_init.html. This site for the SEI's Software Product Line Practice Initiative provides an extensive source of information, including reports, bibliographies, presentations, case studies, and conferences.
www.iese.fhg.de/Business_Areas/Product_Line_Development. The Fraunhofer Institute for Experimental Software Engineering offers many references and links as well as an overview of its Pulse method for software product line development.
P. Clements and L. Northrop, Software Product Lines: Practices and Patterns, Addison-Wesley, 2002. A comprehensive book on software product line engineering practices.
D. Weiss and C. Lai, Software Product-Line Engineering: A Family-Based Software Development Process, Addison-Wesley, 1999. A general description of important concepts plus a detailed process description of the Fast approach.
C. Atkinson et al., Component-based Product Line Engineering with UML, Addison-Wesley, 2002. Illustrates a model-based approach to designing software product lines based on UML and component-based engineering.