A Philosophy of Software Design Framework
A practical framework for managing the fundamental challenge of software engineering: complexity. Apply these principles when designing modules, reviewing APIs, refactoring code, or advising on architecture decisions.
Core Principle
The greatest limitation in writing software is our ability to understand the systems we are creating. Complexity is the enemy: it makes systems hard to understand, hard to modify, and a source of bugs. Evaluate every design decision by asking "Does this increase or decrease the overall complexity of the system?" — the goal is not zero complexity, but minimizing unnecessary complexity and concentrating the necessary kind where it can be managed.
Scoring
Goal: 10/10. When reviewing or creating a design, score it by counting how many of the eight Quick Diagnostic rows it satisfies (≈1.25 points each), then sanity-check against the bands:
- 9-10 — deep modules with interfaces far simpler than implementations; no information leakage (an implementation can change without touching callers); interface comments capture design intent; design improvement is routine. All eight diagnostics pass.
- 6-8 — mostly deep, but one or two leaks, shallow classes, or undocumented abstractions. 5-6 diagnostics pass.
- 3-5 — classitis or temporal decomposition, recurring leakage, comments that only restate code. 2-4 diagnostics pass.
- ≤2 — tactical-tornado code: shallow modules, pervasive leakage, no design intent recorded. 0-1 diagnostics pass.
Always state the current score, the diagnostic rows that failed, and the specific change each one needs to reach 10/10.
The Software Design Framework
Six principles for managing complexity and producing systems that are easy to understand and modify:
1. Complexity and Its Causes
Core concept: Complexity is anything about a system's structure that makes it hard to understand and modify. It shows three symptoms — change amplification, cognitive load, and unknown unknowns — and has two causes: dependencies and obscurity.
Key insights:
- Change amplification: a simple change requires edits in many places
- Cognitive load: a developer must hold too much in mind to make a change
- Unknown unknowns: it isn't obvious what must change or what information is relevant — the worst symptom
- Complexity is incremental — it accumulates from hundreds of small decisions ("death by a thousand cuts"), so every decision matters
Code applications:
| Context | Pattern | Example |
|---|---|---|
| Change amplification | Centralize shared knowledge | Extract color constants instead of hardcoding #ff0000 in 20 files |
| Cognitive load | Reduce what developers must know | open(path) instead of requiring buffer size, encoding, lock mode |
| Unknown unknowns | Make dependencies explicit | Type systems and interfaces surface what a change affects |
| Obscurity | Name things precisely | numBytesReceived not n; retryDelayMs not delay |
See references/complexity-symptoms.md when you need to name which symptom a codebase has before fixing it — per-symptom recognition tests, the dependency taxonomy (syntactic/semantic/temporal/hidden), the C = Σ(cp·tp) cost formula, and a 10-row red-flag table.
2. Deep vs Shallow Modules
Core concept: The best modules are deep: powerful functionality behind a simple interface. Shallow modules have complex interfaces relative to the functionality they provide — they add complexity rather than hiding it.
Why it works: The interface is the cost a module imposes on the rest of the system; the implementation is the benefit. So a method that is harder to learn than to re-implement yourself is net-negative — depth, not line count, decides whether a module earns its place.
Key insights:
- Depth = functionality provided / interface complexity imposed (Unix file I/O is deep; thin Java I/O wrappers are shallow)
- "Classitis": the disease of creating too many small, shallow classes — each interface adds cognitive load
- Small methods are not inherently good; depth matters more than size
- The best abstractions hide significant complexity behind a few simple concepts
Code applications:
| Context | Pattern | Example |
|---|---|---|
| Deep module | Hide complexity behind simple API | file.read(path) hides disk blocks, caching, buffering, encoding |
| Classitis cure | Merge related shallow classes | RequestParser + RequestValidator + RequestProcessor → one RequestHandler |
| Interface simplicity | Fewer parameters, fewer methods | config.get(key) with sensible defaults, not 15 constructor parameters |
See references/deep-modules.md when judging whether an abstraction pulls its weight — before/after code for the depth ratio, the classitis cure worked out, and case studies (Unix I/O, GC, TCP/IP).
3. Information Hiding and Leakage
Core concept: Each module should encapsulate knowledge not needed by other modules. Information leakage — one design decision reflected in multiple modules — is one of the most important red flags in software design.
Why it works: A decision that lives in one module can change there and nowhere else; the same decision leaked into N modules turns one edit into N edits that no compiler will remind you to make. Hiding is what converts change amplification back into a local change.
Key insights:
- Temporal decomposition causes leakage: splitting code by when things happen forces shared knowledge across phases — organize by knowledge instead
- Back-door leakage through data formats, protocols, or shared assumptions is the subtlest form
- Decorators frequently leak — they expose the decorated interface
- If two modules share knowledge, merge them or create a new module that encapsulates it
Code applications:
| Context | Pattern | Example |
|---|---|---|
| Format leakage | Centralize serialization | One module owns JSON encoding/decoding, not json.dumps everywhere |
| Temporal decomposition | Organize by knowledge, not time | Combine "read config" and "apply config" into one config module |
| Protocol leakage | Abstract transport details | MessageBus.send(event) hides HTTP vs. gRPC vs. queue |
See references/information-hiding.md when a change forces you to edit two modules in lockstep — the four leakage forms with code (interface, back-door, temporal, decorator), five reduction strategies, the HTTP-handling case study, and a detection table.
4. General-Purpose vs Special-Purpose Modules
Core concept: Design modules that are "somewhat general-purpose": an interface general enough to support multiple uses, with an implementation that handles current needs. Ask: "What is the simplest interface that will cover all my current needs?"
Why it works: Counterintuitively, the general interface is usually the simpler one — special-case methods multiply as requirements grow, while one general method absorbs them. The trap is the other direction: generality the current needs don't demand is speculative complexity, paid now for a use case that may never arrive.
Key insights:
- "Somewhat general-purpose" is the sweet spot between too specific and too generic
- Push complexity downward: lower-level modules should handle hard cases so upper levels stay simple
- Configuration parameters often represent a failure to decide — each parameter is complexity pushed onto the caller
- When in doubt, implement the simpler, more general-purpose approach first
Code applications:
| Context | Pattern | Example |
|---|---|---|
| API generality | Design for the concept, not one use case | text.insert(position, string) instead of text.addBulletPoint() |
| Reduce configuration | Determine behavior automatically | Auto-detect file encoding instead of an encoding parameter |
| Avoid over-specialization | One general method over many specific ones | store(key, value, options) instead of storeUser(), storeProduct(), storeOrder() |
See references/general-vs-special.md when choosing how general an interface should be — the "simplest interface for all current needs" test, the configuration-parameter antipattern, and push-complexity-downward worked through.
5. Comments as Design Documentation
Core concept: Comments should describe what is not obvious from the code: design intent, abstraction rationale, invariants, and assumptions. "Good code is self-documenting" is a myth for anything beyond low-level implementation detail.
Why it works: Code can only ever record what it does — never why this approach over the alternatives, or what it silently assumes. That rationale is the most perishable information in a system: it lives only in the author's head and is gone the moment they move on, so a comment is the single chance to capture it.
Key insights:
- Four types: interface comments (most important — they define the abstraction), data structure member comments, implementation comments, cross-module comments
- Write comments first (comment-driven design) to clarify thinking before code
- Don't repeat what the code makes clear; keep comments next to the code they describe and update them together
- If a comment is hard to write, the design may be too complex
Code applications:
| Context | Pattern | Example |
|---|---|---|
| Interface comment | Describe the abstraction, not the implementation | "Returns the widget closest to position, or null if none within threshold" |
| Data structure comment | Explain invariants | "List is sorted by priority descending; ties broken by insertion order" |
| Implementation comment | Explain why, not what | "// Binary search: list is always sorted, can hold 100k+ items" |
| Cross-module comment | Link related decisions | "// This timeout must match the retry interval in RetryPolicy.java" |
See references/comments-as-design.md when writing or reviewing comments and unsure what belongs in one — the four comment types with examples, the comment-driven-design procedure, and the rebuttal to the self-documenting-code myth.
6. Strategic vs Tactical Programming
Core concept: Tactical programming gets features working quickly and accumulates complexity with each shortcut. Strategic programming invests 10-20% extra effort in good design, treating every change as an opportunity to improve structure.
Why it works: Tactical speed is borrowed: each shortcut makes future changes harder, while the strategic investment compounds — strategically designed systems are faster to work with within months.
Key insights:
- Tactical tornado: a developer who ships fast but leaves wreckage — celebrated short-term, destructive long-term
- Your primary job is a great design that happens to work, not working code that happens to have a design
- Startups need strategic programming most — early shortcuts compound into crippling debt as the team grows
- Every change is an investment opportunity: leave the code a little better; refactoring is part of every feature, not a special event
Code applications:
| Context | Pattern | Example |
|---|---|---|
| Tactical trap | Resist quick-and-dirty fixes | Don't add a boolean parameter for "just this one special case" |
| Strategic investment | Improve structure during feature work | Refactor an awkward module interface while adding the feature |
| Design reviews | Evaluate structure, not just correctness | Ask "does this make the system simpler?" not just "does it work?" |
See references/strategic-programming.md when deciding how much design effort a change deserves, or making the case for it — the 10-20% investment math, the tactical-tornado pattern, and why startups need strategic programming most.
Common Mistakes
| Mistake | Why It Fails | Fix |
|---|---|---|
| Creating too many small classes | Classitis adds interfaces without depth; each boundary is cognitive overhead | Merge related shallow classes into deeper modules |
| Splitting modules by temporal order | "Read, then process, then write" forces shared knowledge across modules | Group code that shares knowledge into one module |
| Exposing implementation in interfaces | Callers depend on internals; changes propagate | Design interfaces around abstractions; hide formats and protocols |
| Treating comments as optional | Design intent and assumptions are lost; newcomers guess wrong | Write interface comments first; maintain with the code |
| Configuration parameters for everything | A parameter offloaded to the caller is a decision you declined to make (see §4) | Determine behavior automatically; provide sensible defaults |
| Quick-and-dirty tactical fixes | Shortcuts compound until the system is unworkable | Invest 10-20% extra; treat every change as a design opportunity |
| Pass-through methods | A method that only forwards its arguments to another adds an interface but no functionality | Merge the pass-through into the caller or the callee |
| Designing for specific use cases | Special-purpose interfaces accumulate special cases | Ask: simplest interface covering all current needs? |
Quick Diagnostic
| Question | If No | Action |
|---|---|---|
| Can you describe each module in one sentence? | Modules do too much or lack purpose | Split into coherent, describable responsibilities |
| Are interfaces simpler than implementations? | Modules are shallow — complexity leaks outward | Hide more; merge shallow classes into deeper ones |
| Can you change an implementation without affecting callers? | Information is leaking across boundaries | Encapsulate the leaked knowledge in one module |
| Do interface comments describe the abstraction? | Design intent lost; module will be misused | Document what the module promises, not how it works |
| Is design discussion part of code reviews? | Reviews catch bugs but not complexity growth | Add "does this reduce complexity?" to review criteria |
| Does each module hide an important design decision? | Modules organized around code, not information | Reorganize so each module owns specific knowledge |
| Can a newcomer understand module boundaries without reading implementations? | Abstractions undocumented or leaky | Improve interface comments; simplify interfaces |
| Are you spending 10-20% of time on design improvement? | Debt accumulates with every feature | Include design improvement in every PR |
Further Reading
For the complete methodology with detailed examples:
- "A Philosophy of Software Design" by John Ousterhout (2nd edition)
About the Author
John Ousterhout is the Bosack Lerner Professor of Computer Science at Stanford and the creator of the Tcl scripting language and Tk toolkit. He developed A Philosophy of Software Design from his Stanford CS 190 course, distilling decades of systems-building experience into principles that apply across languages and scales.