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.