Software Evolution: Nature and Philosophy

Key Insight

The initial development and subsequent modification of software are inevitable aspects of its lifecycle, contrary to the myth that a well-managed project proceeds linearly from stable requirements to code, tested once and then only modified during a 'maintenance phase' after initial delivery. In reality, all successful software undergoes significant changes; coding, debugging, and unit testing alone consume 30-65 percent of typical project effort. Requirements themselves are dynamic, typically changing by 1 to 4 percent per month, which invariably necessitates substantial code alterations. Modern development practices, focusing more on code-centric approaches, embrace this evolutionary nature, leading to even greater expected code changes over a project's lifespan.

Software evolution, much like biological evolution, can either improve quality through beneficial 'mutations' or degrade it into a 'deevolutionary spiral' if modifications are handled poorly. The key distinction lies in whether changes enhance or compromise program quality. Fixing errors with 'logical duct tape' rather than treating modifications as opportunities to refine the original design leads to degradation, serving as a critical warning sign. Furthermore, the context of changes differs significantly between the construction phase and the maintenance phase: construction changes are usually made by the original developers, before the system is live and forgotten, implying lower pressure and a greater allowance for more 'freewheeling' adjustments with reduced penalty for errors compared to post-delivery maintenance.

A common weakness in programming is the unselfconscious approach to software evolution. Recognizing that evolution during development is an important and unavoidable phenomenon allows it to be strategically planned for and leveraged. The guiding principle is the 'Cardinal Rule of Software Evolution,' which states that every modification should improve the program's internal quality. Programmers never possess as much knowledge at the start as they do after writing code; therefore, each opportunity to revise should be used to apply this learned understanding for improvement. Both initial coding and subsequent changes must be undertaken with the explicit intention of facilitating further modifications, moving the code closer to perfection rather than allowing it to decay.