ICSE 2025: CLDK Tutorial

[rahlk]

🔍 CLDK Deep Dive @ ICSE 2025

This discussion accompanies the tutorial session on codellm-devkit (CLDK), a framework that connects traditional static analysis tools with LLMs specialized for code. In this session, we’ll dive into:
• The Java analysis engine powered by JavaParser and WALA
• How we use Maven/Gradle to resolve builds for accurate type analysis
• Generating analysis.json: a language-neutral symbol table and call graph
• Using the Python SDK to parse, enrich, and query these insights with simple APIs

Feel free to use this thread to ask questions, share ideas, or follow up on examples from the session!

[CalixBarrus]

Where does the model get specified? (Have we not gotten to that part of the tutorial?)

[CalixBarrus]

Nevermind the question was answered!

[rahlk]

Great! Just for reference, it was here—

from openai import OpenAI


def prompt(message: str) -> str:
    """
    Function to prompt the user for input.
    """
    client = OpenAI(
        base_url="https://openrouter.ai/api/v1",
        api_key=os.getenv("OPENROUTER_API"),  # OpenRouter API key
    )
    completion = client.chat.completions.create(
        model="meta-llama/llama-3.2-3b-instruct:free", messages=[{"role": "user", "content": message}]
    )

    return completion.choices[0].message.content

[jff]

Is it possible to see the CFG and other artifacts that were built? Where are they? Thanks

[rahlk]

Absolutely! There are two ways:

1. Fastest way is by using the flag cldk.analysis(..., analysis_json_path='/path/to/artifacts/'). This will create a file called /path/to/artifacts/analysis.json which will have all the data.

2. You could also run codeanalyzer-java as a standalone CLI tool. E.g.,

# Download the latest version
wget https://github.com/codellm-devkit/codeanalyzer-java/releases/download/v2.3.3/codeanalyzer-2.3.3.jar 

# Run code analyzer with Java (11+)
java -jar /path/to/codeanalyzer-2.3.3.jar --input /your/java/project --output /path/to/save/artifacts/ --verbose --analysis-level 2

This will also create a file called /path/to/save/artifacts/analysis.json which will have all the data.