It Will Never Work in Theory

Review: The Essence of Software Engineering

Posted Aug 12, 2013 by Jorge Aranda and Greg Wilson

| Reviews |

Ivar Jacobson, Pan-Wei Ng, Paul E. McMahon, Ian Spence, and Svante Lidman: The Essence of Software Engineering: Applying the SEMAT Kernel. Addison-Wesley Professional, 2013, 978-0321885951.

SEMAT started almost four years ago as a call to action to refound software engineering on a strong theoretical and empirical basis, and to move beyond fads and petty methodological wars to embrace a "kernel" of widely agreed elements. It had an impressive list of signatories, and SEMAT contributors have since sent a submission to the OMG describing what they think the essence of software engineering is, which they hope will become an Object Management Group (OMG) standard this year.

The Essence of Software Engineering, by Jacobson, Ng, McMahon, Spence, and Lidman, is meant to be a practitioner-oriented book describing the SEMAT kernel. Unfortunately, the "kernel" described in the book throws all of the SEMAT goals out of the window, and instead serves as a perfect example of everything SEMAT was supposed to fix.

First, where SEMAT wanted to build upon a strong theoretical foundation, this kernel is not informed by any established theories from any discipline. It does not even articulate any theories of software engineering, but instead claims that that will be handled separately by the Theory Working Group. (Tellingly, the Theory Track page in the SEMAT website is currently blank.)

Second, where SEMAT wanted to build upon "credible experimental evaluation", this kernel does not seem to be informed by any empirical findings. Instead, the only "evidence" put forward to back up its recommendation are the anecdotal experiences of its authors. At a time when empirical software engineering is stronger than ever, and its links to developments in other fields are strengthening and being properly acknowledged, it is frankly nonsensical for someone to claim that they have distilled software engineering's "essence" while completely ignoring the hundreds of empirical studies that have been conducted in the past two decades.

Finally, and perhaps most damningly, where SEMAT's stated goal was to move away from fads, this kernel promotes a methodology developed by a single consulting firm (that of the book's first author), along with idiosyncratic terms like "alphas" and some slideshow-friendly visual aids whose semantics are even less well defined than those of UML 1.0.

But is this "kernel" any good? It's impossible to tell in the absence of any objective evaluation, but the signs aren't hopeful. For a start, it ignores everything that's actually problematic about constructing a universal core for software development. Anything deserving of that name should work for Microsoft as well as for open source, for IBM as well as for a tiny startup, for grad students writing scripts to process their data as well as for videogame companies, defense contractors, and OS developers. That's simply not the case here.

As an example, SEMAT wants development teams to keep track of several areas of concern—the aforementioned alphas—such as stakeholders, requirements, and systems. At any time, each alpha is in some particular state (e.g., requirements can be "conceived", "bounded", "coherent", "acceptable", "addressed", and finally "fulfilled"), and this state changes as the project progresses. That may sound nice in the boardroom, but as many researchers have reported, this is not the way requirements progress for many successful software groups. Under some conditions requirements can be exploratory and unarticulated, and projects can still succeed—not despite this fluidity, but because of it.

What's more, the whole notion of concrete states is flawed. In reality, the state of many parts of a project can be fuzzy, there can be valid disagreement as to what state something is in, and disagreement as well about what it means to be in one state or another [1]. Essence's "solution"? Have everyone in the room take an educated guess, and move on if there is consensus. If there isn't, well, then maybe some people in the room have not properly understood each state, so they should spend some time understanding the states better and taking another vote [2].

To sum up, if this book actually represents the SEMAT initiative's views, then that group has managed to get some well-known names, and possibly the OMG, to endorse yet another fad that ignores what we actually know about the complexity and diversity of actual software development worldwide. Today's empirical studies of software engineering are imperfect and incomplete, but they're a more substantial foundation for real engineering than anything you'll find in this book.

[1] James C. Scott's explores this idea in considerable depth in his insightful book Seeing Like a State (Yale University Press, 1999, 978-0300078152). He shows that centralizers have frequently preferred manageable uniformity to local custom and adaptation, even when the latter is more productive and more robust. Essence's alphas are unlikely to cause as much suffering as the forced collectivization Scott analyzes, but they are unlikely to produce any better results.

[2] And people not present in the room where alpha states are being determined should presumably learn not to miss meetings...

Comments powered by Disqus