The Software Engineering Body of Knowledge (SWEBOK) states that software engineering transforms problems stated in natural language to problems stated in electrical current. This transform view emphasizes the continuity of the design process and highlights intermediate artifacts.
Update 2022-01-14: Here’s SWEBOK term definitions if needed
Abstractly speaking, problem solving using a computer can be considered as a process of problem transformation - in other words, the step-by-step transformation of a problem statement into a problem solution. To the discipline of software engineering, the ultimate objective of problem solving is to transform a problem expressed in natural language into electrons running around a circuit. - SWEBOK 13.1.5
The major steps of this process are
- Figure out what to tell the computer to do (requirements)
- Convert the problem statement into algorithms (code)
- Convert the algorithms into machine instructions (compile)
Compilers Make Machine Code
I don’t know about you, but I never find myself manually assembling code into machine instructions. That’s what compilers are for, and they’re good at their job. The entirety of a typical software process lives in those two first steps, defining the problem and translating it into rigorous code.
In fact, Jack Reeves has a famous article about this very topic called What is Software Design?.
Everything else is design
Software is great at automation. That includes our own programming processes. Any mechanical steps, like compilation, get automated away.
This means no part of the development process is a prescriptive or mechanical assembly process. Every stage is related to further defining and expressing the problem being solved. The initial real-world processes tend to be messy. They can’t always be mapped into precise algorithmic processes, or the mapping may not be economical. Many such translation challenges are not apparent until detailed design or even until the code is written.
This means there design question all the way through, so the original problem must be understood all the way through or translations end up like repeated google translations. Developers at all phases have to work with stakeholders to evaluate trade-offs and approximations.
SWEBOK Section 13.3.1 elaborates on the breakdown of SWEBOK itself based on this transformation view. It frames typical development phases (and book chapters) as transformations creating a problem statement one step closer to machine code.
Consequently, this view highlights artifacts that each stage produces. It also underscores an important quality measure: each artifact should communicate effectively the same knowledge, because each is a restatement of the same problem. In turn, it underscores the importance of understandable artifacts and reflection of the domain in code (see DDD).
Relation to common process
Common process mostly differ in how frequently they complete this transformation pipeline, the size of work translated at one time, and formality of stage artifacts.
- Kanban maintains a constant loop based on tasks.
- Scrum partitions cycles into short fixed-time increments (~1-2 weeks)
- Phased delivery methods break work into larger deliverables (on the scale of months)
- Waterfall attempts to address each transform only once for the whole project
Agile methods are popular because tight loops show incremental progress, reduce need for artifact formality, and improve opportunities for process control. Process loop lengths are also not completely exclusive. It is common in many methods to plan rigorously at the scale of weeks, and less specifically over the course of a quarter, and thematically over several years.
There are many views of software process that capture different truths. I think the SWEBOK transform view captures the nature of software as a problem-clarification process. Every step must understand the original problem and handle messy design trade-offs. It also clarifies that each stage artifact should communicate the same fundamental problem, just in progressively more detail.