anastasiyaw/codex-claude-code-config/code-complexity

Code complexity — keeping a unit comprehensible

Complexity is not how clever the code is. It is how much you must hold in your head to change one line safely. It accumulates by increments that each look acceptable, which is why it is never one bad commit's fault and always everyone's problem.

The two symptoms worth memorising

Change amplification — one decision, many edits. If changing a timeout means editing five files, that decision was not encapsulated anywhere.

Cognitive load — how much you must know to be safe. A function that is correct only if you know that the caller already took a lock has exported its complexity to everyone who reads it.

Both are properties of the interface, not the implementation. Which gives the rule:

Deep over shallow

A module's value is functionality it provides ÷ interface you must learn. Deep = small interface, substantial behaviour behind it. Shallow = large interface, little behind it.

• A pass-through method that only calls one other method with the same arguments is

negative value: it adds an interface without adding behaviour.

• "One class per concept, many tiny classes" is not automatically good. Many shallow

classes cost more total interface than a few deep ones.

• The right question is never "is this class small?" but "how much do I need to know to

use it?"

Information hiding — and its inverse

Every design decision that might change should be known to exactly one module. The inverse — information leakage — is when a decision shows up in two places: a file format known to both the reader and the writer, a status string parsed by three modules, a lock the caller has to remember.

Test: if this decision changed, how many files would I edit? More than one is leakage.

Names, functions, errors — the unit level

• Names: a name that needs a comment to explain it is the wrong name. If you cannot

name it, you do not yet know what it does — that is information, not a naming problem.

• Functions: one job, one level of abstraction, few parameters. Three or more

booleans in a signature is a sign that several functions are hiding in one.

• Comments: comment the why and the constraint, never the what. A comment that

restates the code rots the moment the code changes. A comment that records a ceiling or a reason is the most durable thing in the file.

• Errors: define them out of existence where you can (an operation that cannot fail

needs no error path); where you cannot, fail loudly and early. Silently swallowing is the failure that looks like success — the most expensive kind.

• Tests: one assertion of behaviour per test, named for the behaviour. A test that

needs a comment to say what it proves is not proving it clearly.

The pragmatic meta-rules

• DRY is about knowledge, not text. Two identical lines expressing two unrelated

decisions are not duplication. One decision expressed in two places is, even if the code looks different.

• Orthogonality: changing one thing should not move another. If it does, they are

coupled through something you have not named.

• Design by contract: state preconditions and invariants, and assert them. An

assertion is executable documentation that cannot rot.

• Broken windows: one tolerated mess licenses the next. Fix it while it is one.
• Tracer bullets: build one thin end-to-end path first, then widen. It proves the

seams; a stack of half-built layers proves nothing.

• Reversibility: prefer decisions that can be undone. There are fewer final answers

than the design meeting suggests.

Review pass — in this order

1. Does any interface require knowledge it did not give you? (leakage)
2. Would a change to one decision edit more than one file? (amplification)
3. Any pass-through methods, or classes whose interface is most of their content?
4. Any name that needs its comment?
5. Any error path that continues on failure without saying so?
6. Any duplicated decision — not duplicated text?

References — load on demand

• references/software-design-philosophy/deep-modules.md, information-hiding.md,

complexity-symptoms.md, general-vs-special.md, strategic-programming.md, comments-as-design.md

• references/clean-code/naming-conventions.md, functions-and-methods.md,

error-handling.md, testing-principles.md, comments-formatting.md, code-smells.md

• references/pragmatic-programmer/dry-orthogonality.md, contracts-assertions.md,

tracer-bullets.md, reversibility.md, broken-windows.md, estimation-portfolio.md

• *-original.md — the source skills' own framework prose, kept verbatim

Gotchas

• Small is not simple. Splitting a deep module into six shallow ones raises total

interface and lowers comprehensibility while looking like progress.

• Abstraction without a second caller. An interface with one implementation and no

test double is a guess about the future priced as complexity today.

• DRY applied to text. Merging two coincidentally-identical blocks couples two things

that had no reason to change together, and the next change breaks one of them.

• "Clean" as a rewrite. Restructuring without a behaviour test is not cleaning, it is

gambling — see refactoring-safely for the transformation with a net.

Troubleshooting