Lessons Learned from Implementing a Language-Agnostic Dependency Graph Parser

In software engineering, automated tools are essential for detecting policy violations within code. These tools typically analyze the relationships and dependencies between components in large codebases, which may be written in various programming languages. Most available tools, whether free or proprietary, rely on third-party software to perform statistical analyses. This approach often requires a separate tool for each programming language, which can lead to high maintenance efforts, and even relying on a standardized technology such as Language Servers has several drawbacks. This paper investigates the feasibility of removing language-specific dependencies in the construction of dependency graphs by using two libraries: Tree Sitter and Stack Graph. After analyzing the capabilities of these technologies, their application in this context is demonstrated, and the effectiveness and accuracy of the proposed solution are evaluated.