17 — Design & Architecture Smells¶

Driving question: Can a smell detector distinguish “unusual” from “harmful”?

Learning objectives¶

Recognize common code, design, and architecture smells.
Interpret threshold-based detectors critically.
Compare structural, historical, and semantic evidence.
Prioritize smells by maintenance risk rather than count.

Smell levels¶

Code smells: Long Method, Feature Envy, Long Parameter List, Duplicated Code.
Design smells: God Class, cyclic dependencies, unstable abstractions, inappropriate inheritance.
Architecture smells: hub-like dependency, cyclic architecture, scattered functionality, ambiguous interfaces.

Detection model¶

A simple metric-based detector might define God Class as:

WMC > t1 AND ATFD > t2 AND TCC < t3

The research questions are immediately visible:

How were thresholds chosen?
Do they generalize across projects/languages?
What is the oracle/ground truth?
How much do experts agree?
Does detected smell removal improve outcomes?

Prioritization¶

Severity can incorporate change frequency, defect history, ownership, centrality, test coverage, and refactoring cost—not only smell intensity.

Design / research exercise¶

Run two smell detectors or two threshold settings on the same project. Analyze disagreements manually. Classify disagreements as threshold effects, parser/modeling differences, or conceptual ambiguity.