IEEE Software, Vol. 15, No. 1, January/February 1998
The Art, Science, and Engineering of Software Development
When interviewing candidates for programming jobs, one of my favorite interview
questions is, "How would you characterize your approach to software
development?" I give them examples such as carpenter, fire fighter, architect,
artist, author, explorer, scientist, and archeologist, and I invite them to come up with
their own answers. Some candidates try to second-guess what I want to hear; they usually
tell me they see themselves as "scientists." Hot-shot coders tell me they see
themselves as commandos or swat-team members. My favorite answer came from a candidate who
said, "During software design, Im an architect. While Im designing the
user interface, Im an artist. During construction, Im a craftsman. And during
unit testing, Im one mean son of a bitch!"
I like to pose this question because it gets at a fundamental issue in our field: What
is the best way to think of software development? Is it science? Is it art? Is it craft?
Is it something else entirely?
Two cultures or an
We have a long tradition in the software field of debating whether computer programming
is art or science. Thirty years ago, Donald Knuth began writing a seven-volume series, The
Art of Computer Programming. The first three volumes stand at 2,200 pages, suggesting
the full seven might amount to more than 5,000 pages. If thats what the art of
computer programming looks like, Im not sure I want to see the science!
People who advocate programming as art point to the aesthetic aspects of software
development and argue that science does not allow for such inspiration and creative
freedom. People who advocate programming as science point to many programs high
error rates and argue that such low reliability is intolerablecreative freedom be
damned. In my view, both these views are incomplete and both ask the wrong question.
Software development is art. It is science. It is craft, fire fighting,
archeology, and a host of other activities. It is as many different things as there are
different people programming. But the proper question is not "What is software
development?" but rather "What should software development be?" In
my opinion, the answer to that question is clear: Software development should be engineering.
Is it? No. Should it be? Unquestionably, yes.
The dictionary definition of engineering is the application of scientific and
mathematical principles toward practical ends. That is what most of us try to do,
isnt it? We apply scientifically developed and mathematically defined algorithms,
functional design methods, quality-assurance practices, and other practices to develop
software products and services. As David Parnas points out, in other technical fields the
engineering professions were invented and given legal standing so that customers could
know who was qualified to build technical products ("Software Engineering: An
Unconsummated Marriage," Software Engineering Notes, Nov. 1997). Software
customers deserve no less.
Some people object to the idea that software development should be treated as
engineering because they think "software engineering" is just a buzzword; they
argue that no core body of knowledge can be identified as "software
engineering." Thirty years ago when the first NATO conference on software engineering
was held, this statement was undoubtedly true. The first paper on structured design had
not yet been published. Neither had the first papers on inspections, measurement-based
estimation, or the high performance variability among individual programmers. Ten years
would pass before the publication of the first books on structured design, software system
specification, and metrics. Online systems were controversial, and the first readily
available GUI interfaces were still more than a decade away. Whatever core body of
knowledge could have been defined at that time would have been at least 50 percent
obsolete within 10 years.
Today, "software engineering" is still thrown around as a buzzword more often
than not. Thats unfortunate. But the fact that the term is abused does not mean it
has no legitimate meaning. Software development has come a long way in 30 years. We still
do not have an absolutely stable core body of knowledge, and knowledge related to specific
technologies will never be very stable, but we do have a body of knowledge that is stable
enough to call software engineering. That core includes practices used in requirements
development, functional design, code construction, integration, project estimation,
costbenefit trade-off analysis, and quality assurance of all the rest.
Many core elements have not yet been brought together in practically oriented textbooks
or courses, and in that sense our body of knowledge is still fragmented and under
construction. But the basic knowledge about how to perform each of these practices is
availablein journal articles, conference papers, and seminars. The pioneers of
software engineering have already blazed the trails and surveyed the land. Now the
software engineering settlers need to get to work, turning the trails into roads and
developing the rest of the education and accreditation infrastructure.
Some people think that treating software development as engineering means well
all have to use formal methods, which both common sense and experience tell us are
overkill for many projects. Others say that commercial software is too dependent on
changing market conditions to permit careful, time-consuming engineering. These objections
are based upon narrow, and I think, mistaken, ideas of engineering. Engineering is the
application of scientific principles toward practical ends. If the engineering
isnt practical, it isnt good engineering.
Treating software as engineering makes clearer the idea that different development
goals are appropriate for different projects. When a building is designed, the
construction materials must suit the buildings purpose. I can build a large
equipment shed to store farming vehicles from thin, uninsulated sheet metal. I
wouldnt build a house the same way. But even though the house is sturdier, warmer,
and likely to last longer, we dont refer to the shed as having lower quality than
the house. The shed has been designed appropriately for its intended purpose. If it had
been built as robustly as a house, we might even criticize it for being
"overengineered"a judgment that the designers wasted resources in building
Similarly, each software program must strike a balance between functionality and
reliability appropriate to its purpose. We dont want to pay $5,000 for a word
processor, nor do we want one that crashes every 15 minutes. Thus, the shrink-wrap
software industry is constantly optimizing an equation that includes the variables of time
to market, reliability, functionality, and cost.
In most cases, these optimizations are business decisions rather than moral decisions.
There is no inherent "right" or "wrong" quality level for software
independent of the specific software package being created. The "right"
materials depend on a specific buildings purpose; the right reliability depends on
the specific softwares purpose. When safety is involved, the optimizations become
moral decisions, but such calculations are not unique to software. Drug manufacturers
weigh the benefits of their drugs against their side effects. Bridge designers and
aeronautical engineers consider both safety and cost. Whether the calculations involve
functionality, time, money, or human safety, striking the right balance in a deliberate,
informed way is properly the domain of engineeringapplying scientific principles
toward practical ends. Many current software projects strike that balance poorly, or with
little awareness that trade-offs even exist. An engineering approach can only help them.
Roles in Software
Another objection is that treating software development as engineering will take all
the fun out of it. Software development in many quarters is a fast-and-loose craft, and
many fast-and-loose craftsmen are making fantastic incomes without knowing much about
effective development practices. Only a prophet could know how long the public will
tolerate computer programmers who make doctors salaries without comparable training
and education, but I predict that one fallout of year 2000 problems will be public
insistence on regulating software developers.
All other engineering fields are self-accredited and self-regulated; software
developers will either voluntarily swallow the medicine of software as engineering or we
will have it forced down our throats. I know which way I prefer to take my medicine.
As roles in the field settle out, I wont be surprised to see greater coordination
between software engineering and other engineering disciplines. Some students will major
in electrical engineering or aeronautical engineering with an emphasis in software. They
will lead the software parts of their engineering projects, and their classmates who
choose pure electrical engineering or aeronautical engineering will not. Some students
will attain degrees in general software engineering.
A major distinction between software engineering and other kinds of engineering is that
software is so labor intensive that a significant amount of engineering energy must be
focused on project goals in addition to product goalson the means to
the ends as well as the ends themselves. Other kinds of engineers choose components to be
used in the structures they build. Software engineers choose both the tools used to build
the software and the components to be used within the software itself.
The fact that some software must be produced by trained, accredited engineers
doesnt imply the same for all software. I can fix a broken railing on my stairs
without a civil engineering degree. I can mix a gin and tonic without a chemical
engineering degree. But can I build a second story on my house? No. My local government
will require me to have an engineer sign the building plans. We should eventually see a
similar stratification of software jobs into amateurs, skilled craftsmen, and engineers.
Once we stop asking the trick question of, "Is software development
engineering?" and start asking the real question of, "Should software
development be engineering?" we can start answering really important questions: What
is software engineerings core body of knowledge? How should software engineers be
trained? How should software engineers be certified? And, perhaps the hardest to answer,
How long will it take for all this to happen?
Editor: Steve McConnell, Construx Software, 11820 Northup
#E200, Bellevue, WA 98005.
E-mail: firstname.lastname@example.org - WWW: http://www.construx.com/stevemcc/