CRAB Webapp (Code Review Automation Benchmark)

A research-driven platform for evaluating deep learning models on automated code review tasks. CRAB provides two core services:

• Dataset download: Obtain high-quality, curated Java code review datasets for comment generation and code refinement tasks.
• Result evaluation: Upload model-generated predictions to receive standardized evaluation metrics via a REST+WebSocket API.

Table of Contents

• Features

• Prerequisites

• Installation & Setup

  • Environment Variables

• Running the Application

• Using the Webapp

  • Download a Dataset
  • Upload Predictions
  • Track Submission Status

• API Endpoints

• Project Structure

• Contributing

• License

• Acknowledgements

Features

• Static Frontend: Vanilla HTML/CSS/JS interface—no build toolchain required.
• Dataset Delivery: ZIP archives of JSON files, with optional full repo context.
• Submission Queue: Server-managed job queue with configurable parallelism (via MAX_WORKERS).
• Real‑time Feedback: Progress updates over WebSockets (using Flask-SocketIO).
• Robust Data Processing: Utilities for parsing, validating, and evaluating submissions in src/utils.

Prerequisites

• Python 3.8+
• (Optional) Docker daemon if you wish to execute the code refinement evaluation

Installation & Setup

1. Clone the repository:

   git clone https://github.com/karma-riuk/crab-webapp.git
   cd crab-webapp
   

2. Install Python dependencies:

   pip install -r requirements.txt
   

Environment Variables

Defaults are set in src/utils/env_defaults.py (port 45003, data/ path, etc.) To override:

cp .env.example .env
# Edit .env to adjust:
# PORT=..., MAX_WORKERS=..., DATA_PATH=..., RESULTS_DIR=...

Running the Application

From the project root:

python src/server.py

• The Flask app serves static files from public/ at / and mounts API routes under /datasets and /answers via Blueprints.
• By default, open your browser to http://localhost:45003/.

Using the Webapp

Download a Dataset

1. Select Comment Generation or Code Refinement.
2. (Optional) Check Include context to get full repo snapshots.
3. Click Download to receive a ZIP with JSON (see schemas in public/index.html)

Upload Predictions

1. Choose task type (comment or refinement).
2. Select your JSON file (matching the dataset schema).
3. Click Upload JSON.
4. The server responds with a process ID.

Track Submission Status

• Progress bar displays real-time percentage via WebSocket events (requires X-Socket-Id for subscribing to updates).

• You can also poll GET /answers/status/<id> to retrieve a simple JSON object:

  • status: created, waiting, processing, or complete

  • Once status is complete, the response includes:

    {
      "type": "comment" | "refinement",
      "results": { /* evaluation metrics or processed data */ }
    }
    

API Endpoints

Method	Route	Description
GET	/datasets/download/<dataset>	Download ZIP of comment_generation or code_refinement (use ?withContext=true for full repo).
POST	/answers/submit/comment	Submit comment-generation JSON.
POST	/answers/submit/refinement	Submit code-refinement JSON.
GET	/answers/status/<id>	Poll status or results (may include X-Socket-Id for notifications).

Project Structure

├── data/                       # Dataset files: dataset.json, archives, etc.
├── public/                     # Static frontend
│   ├── css/style.css           # Styles
│   ├── img/crab.png            # Icon
│   ├── index.html              # UI with modals, schema docs
│   └── js/                     # Frontend scripts
│       ├── index.js            # UI logic, fetch & WebSocket handlers
│       ├── modal.js            # Modal dialogs
│       └── sorttable.js        # Table sorting
├── src/                        # Backend source
│   ├── server.py               # App entry: Flask + SocketIO
│   ├── routes/                 # Blueprints
│   │   ├── index.py            # Root & health-check
│   │   ├── datasets.py         # File downloads
│   │   └── answers.py          # Submission & status endpoints
│   └── utils/                  # Core logic & helpers
│       ├── env_defaults.py     # Default ENV vars
│       ├── dataset.py          # Load/validate dataset JSON
│       ├── process_data.py     # Evaluation functions
│       ├── observer.py         # WebSocket observer & queue cleanup
│       ├── queue_manager.py    # Concurrency control
│       └── build_handlers.py   # Build/test wrappers
├── requirements.txt            # Python libs: Flask, SocketIO, dotenv, etc.
├── TODO.md                     # Next steps and backlog
└── .env.example                # Template for environment variables

Acknowledgements

• Developed as part of a Master's thesis at Università della Svizzera Italiana.