The Laws of Software Evolution

Lehman’s paper is presented as an early account of why software keeps evolving and why software engineering is distinct from programming. The page says he classifies programs into S-, P-, and E-programs, sketches an ideal lifecycle that does not match practice, and names five laws that still fit experience.

Reading notes

#

• Large programs are described as never complete because adding functionality to existing systems often feels easier than starting over.
• Lehman is credited with separating programming from software engineering by arguing for a discipline with insight, theory, models, methodologies, techniques, and tools.
• S-programs map inputs to outputs against a fixed specification and can be fully verified.
• P-programs are defined by business rules and real-world usefulness, but their correctness depends on external criteria.
• E-programs combine P- and S-programs into software that mechanizes a human endeavor or system and keeps adapting over time.
• The ideal development cycle is listed as requirements, top-level design, recursive component design, implementation of the specs, integration, deployment, and then maintenance.
• The page says this ideal cycle does not match practice because specification, coding, integration, and deployment are usually interleaved and overlapped.
• The five laws named on the page are continuing change, increasing complexity, the fundamental law, conservation of stability, and conservation of familiarity.
• Continuing change means software is never done and must adapt to new requirements, environmental changes, and feedback.
• Increasing complexity means software becomes more complex unless work is done to manage that complexity.
• The fundamental law says the system as a whole can be measured and understood statistically even when local decisions are made from limited knowledge.
• Conservation of stability says output rate stays constant, so adding people is needed just to keep the same pace as maintenance grows.
• Conservation of familiarity says growing software becomes more constrained by past choices and large changes become more disruptive as the user base grows.