Unlocking Code‑Level Quality: From Static Analysis to Intelligent Testing

1. Background

Code‑level quality technology aims to improve software quality by operating directly on source code or compiled artifacts. It goes beyond simple static scans or unit tests, targeting code understanding, instrumentation, and automated defect location.

2. Architecture of Code‑Level Quality

The architecture is divided into two major layers. The first layer (CodeC) focuses on code comprehension, providing services such as AST extraction, call‑graph construction, dependency analysis, and risk scoring through storage, analysis, modeling, and API tiers. The second layer (Codeπ) applies the insights to concrete quality goals, including code quality (CodeQ), code probing (CodeP), code health (CodeH), and defect location (CodeDL).

2.1 Code Understanding (CodeC)

Code understanding extracts semantic information from programs. Typical techniques include static analysis (lexical, syntactic, data‑flow, control‑flow analysis), dynamic analysis, and non‑source analysis. With the rise of large language models, AI‑driven code comprehension is also being explored.

2.2 Code Probing (CodeP)

Probes are inserted into the program to collect runtime data such as execution counts, timing, or memory usage. Common tools include ccover, covtool, JaCoCo, and gocov. Probe data enables coverage metrics, performance monitoring, and business instrumentation.

2.3 Coverage Metrics

Line coverage – whether each executable statement has been executed.

Branch coverage – whether each conditional branch has been exercised.

Path coverage – whether every possible execution path has been traversed.

Changed‑line coverage – coverage of code modified since the last release.

These metrics support scenarios such as developer self‑testing, QA gate‑keeping, precise testing, integration testing, baseline upgrades, gray‑box testing, and automated test capability evaluation.

2.4 Intelligent Unit Testing (UT)

Traditional unit tests are written manually, which is costly. Intelligent UT automatically analyzes a target function, generates test data, and produces test code that exercises edge cases. The workflow consists of function selection → code analysis → test data synthesis → test code generation.

2.5 Rule‑Based Static Analysis (SA)

Static analysis scans source code without execution, detecting defects such as null‑pointer dereferences, array out‑of‑bounds, and division‑by‑zero based on predefined rules. It is lightweight and highly automated but depends on up‑to‑date rule sets.

2.6 Isolated (Orphan) Function Detection

Functions that are never called are called orphan functions. Detecting them reduces technical debt and improves maintainability. Three detection strategies are described:

Static analysis – identify functions with zero inbound calls, adjusting for language‑specific constructs.

Dynamic analysis – record call stacks at runtime via probes.

Hybrid approach – combine static and dynamic results for higher accuracy.

3. Outlook

The article provides a high‑level overview of code‑level quality techniques. Future posts will dive deeper into each layer, present implementation details, and share practical experiences from Baidu’s Quality Efficiency Platform.