Friday, December 14, 2007

Timeless Software

Barry Boehm (having read it many times it's pronounced "beam"), "A View of 20th and 21st Century Software Engineering", lists the timeless qualities of a good programmer that have been discovered through the decades. Some of them include: don't neglect the sciences, avoid sequential processing, avoid cowboy programming, what's good for products is good for process, and adaptability trumps repeatability.

The powerpoint for the presentation is also online.

Update: I have notes from this talk but little time to review them. There's three things that stick out in my mind though.

The first is slide 5. This lists the most common causes of project failure. Most of these are not technical but centre around project management and people. I think it can be successfully argued that the other issues, mainly technical problems, are also people problems too, especially things like "lack of executive support".

Slide 9 has a Hegelian view of the progress made in writing software. That is, there is a process of advancement that involves theses, antitheses and syntheses. For example, software as a craft was an overreaction to software being considered analogous to hardware development and agile methods are a reaction (some may say overreaction) to the waterfall methodology.

Thirdly, Slide 30 has what initially looks like a good diagram to help decide how agile your project should be. I had issues with this. Much of my experience with agile projects is that they are actually better planned than most traditionally run projects. I also don't see how the criticality of the defects is appropriate either, as the number of defects is far fewer and less critical in well run agile projects too. Maybe I've never been on a well run planned project.

This made me wonder, why does software development still seem so far removed from science? Or to put in another way, why is it so dependent on personal experience? OO is quite frequently vilified but the reason people supported and continue to do so was because reuse increased and costs decreased - people were just following the numbers. Before OO, it was structured programming. And in contrast, what seemed like good ideas, such as proving the correctness of programs, have not been widely adopted. Barry points out that the reasons, since its inception in the 1970s, was because it did not prevent defects occurring in the specification and it was not scalable. This lead to the ideas of prototyping and RAD which were developed to overcome these perceived failures.

No comments: