Anti-Pattern Resolutor

Local-first, agentic refactoring pipeline for Java that pairs SonarQube findings with LLM-based reasoning to detect smells/anti-patterns and propose file-scoped, behaviour-preserving edits. Changes are gated by compile + tests; public interfaces are preserved; static analysis is re-run for reporting only. Run artefacts (plans, diffs, logs) are persisted for auditability.

Project made in collaboration with UCL and IBM.

What it does

• Interprets rule-based static analysis (SonarQube) as signals, not ground truth.
• Coordinates single-responsibility agents (Scanner -> Strategist -> Transformer -> Reviewer -> Explainer) with a shared context.
• Enforces compile+test acceptance; tests are never modified.
• Java, file scope only (no cross-file/architectural refactors).
• Provider-agnostic LLM layer (e.g., local Ollama; hosted options supported) with externalised prompts.
• Uses a keyed document Trove (definitions, symptoms, safe remedies) for deterministic retrieval.

Requirements

• Python 3.8+
• An LLM backend (e.g., Ollama locally)
• SonarQube access (local) for static analysis
• Git
• Java JDK (11 recommended)
• Maven (3.9.11 recommended)

Installation & Configuration

# 1. Clone
git clone https://github.com/Andrei-Constantin-Programmer/Anti-Pattern-Resolutor.git
cd Anti-Pattern-Resolutor

# 2. Create and activate virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# 3. Install dependencies
python install_requirements.py # pip install -r requirements.txt works on Unix, 
                               # but not Windows due to incompatible libraries.

Optional (Ollama locally):

ollama pull granite3.3:8b
ollama pull nomic-embed-text

Further LangChain configurations can be changed by modifying AntiPattern_Remediator/config/settings.py.

Usage

Prepare coverage candidates

This stage clones repos, runs tests with JaCoCo, and writes a list of files with 100% line coverage to safely target.

Create repos.txt in the repository root:

https://github.com/org/repo-one
https://github.com/org/repo-two

Run:

# From repository root
python jacoco_tool/jacoco_analysis.py --repos repos.txt

# Useful flags:
#   --single-repo https://github.com/user/repo
#   --clone-dir clones
#   --output-dir jacoco_results
#   --force-jacoco
#   --timeout 600
#   --verbose

Outputs:

• Cloned sources under clones/ (default)
• Coverage artefacts and a combined file list under jacoco_results/ (path is printed at the end of the run)

Provide a SonarQube token

Generate a user token in SonarQube (My Account -> Security), then set it as SONARQUBE_TOKEN.

• Docs: https://docs.sonarsource.com/sonarqube-server/latest/user-guide/managing-tokens/#generating-a-token

Unix:

export SONARQUBE_TOKEN="paste-your-token"

Windows PowerShell (temporary):

$env:SONARQUBE_TOKEN = "paste-your-token"

Windows (persist):

setx SONARQUBE_TOKEN "paste-your-token"
# Restart the terminal afterwards

Run SonarQube

python remediate_repos.py
        --token SONARQUBE_TOKEN # If you haven't added a token in the environment

Run the Remediator

python AntiPattern_Remediator/main.py

The pipeline selects 100%-covered files, proposes minimal, behaviour-preserving edits, and gates them behind compile + test.

SonarQube is re-run for reporting.

Plans, diffs, logs, and summaries are written to the run output directory (path shown in the console).

Troubleshooting

• "No coverage results found"
  Ensure mvn -q -DskipTests=false test succeeds in each repo; consider --timeout and --force-jacoco when running the JaCoCo tool.
• Auth errors with SonarQube
  Confirm SONARQUBE_TOKEN is set in your current shell and your SonarQube URL is reachable.
• Java/Maven not found
  Verify JDK 11 and Maven are on PATH.

Acknowledgments

• IBM - For providing technical expertise and computational resources, and mentorship from Dr Amrin Maria Khan and Prof. John McNamara
• University College London (UCL) - For research guidance and academic support, under the supervision of Dr Jens Krinke
• LangChain Community - For the foundational LLM orchestration framework
• Ollama Project - For making LLM deployment accessible and efficient