Code Knowledge Graph test

Prerequisites

• A neo4j database with the code knowledge graph (see instructions below on how to generate)
• Java 21
• Maven 3.9 and up (wrapper is also provided)
• An openai api key

Running the application

1. Startup the neo4j database
2. Set NEO4J_ADDRESS and OPENAI_API_KEY environment variables with the neo4j database address and openai api key respectively
3. Run ./mvnw exec:java

Setting up neo4j

Go to the codebase that you want to generate a code knowledge graph for and add the JQAssistant plugin

<plugin>
    <groupId>com.buschmais.jqassistant</groupId>
    <artifactId>jqassistant-maven-plugin</artifactId>
    <version>${version.jqassistant}</version>
    <executions>
        <execution>
            <goals>
                <goal>scan</goal>
                <goal>analyze</goal>
            </goals>
            <configuration>
                <warnOnSeverity>MINOR</warnOnSeverity>
                <failOnSeverity>MAJOR</failOnSeverity>
            </configuration>
        </execution>
    </executions>
</plugin>

Add .jqassistant.yml to the root director of the project with contents

jqassistant:
  store:
    embedded:
      neo4j-plugins:
        - group-id: org.neo4j.procedure
          artifact-id: apoc-core
          classifier: core
          version: 5.24.0

Run ./mvnw jqassistant:scan and ./mvnw jqassistant:analyze

Finally run ./mvnw jqassistant:server to serve the neo4j server locally

It might be necessary to prune the generated graph a bit to fit the shcema of the database to the context window of the model. A pruner class is added for this purpose but depending on the context size of the model it might or might not be needed to prune it more.