The growing complexity of modern software systems has highlighted the shortcomings of traditional programming analysis techniques, particularly for Software Engineering (SE) tasks. While machine learning and Large Language Models (LLMs) offer promising solutions, their effectiveness is limited by the way they interpret data. Unlike natural language, source code meaning is defined less by token adjacency and more by complex, long-range, and structural relationships and dependencies. This limitation is especially pronounced for C and C++, where flatter syntactic hierarchies, pointer aliasing, multi-level indirection, typedef-based type obfuscation, and function-pointer calls hinder accurate static analysis. To address these challenges, this paper introduces ATLAS, a Python-based Command-Line Interface (CLI) that (i) generates statement-level Control Flow Graphs (CFG) and type-aware Data Flow Graphs (DFG) that capture inter-functional dependencies for the entire program; (ii) has the ability to work on entire C and C++ projects comprising multiple files; (iii) works on both compilable and non-compilable code and (iv) produces a unified multi-view code representation using Abstract Syntax Trees (AST), CFG and DFG. By preserving essential structural and semantic information, ATLAS provides a practical foundation for improving downstream SE and machine-learning-based program understanding. Video demonstration: https://youtu.be/RACWQe5ELwY Tool repository: https://github.com/jaid-monwar/ATLAS-code-representation-tool

翻译：暂无翻译