Software Model Generation and Simulation for Change Impact Analysis with LLM

Software change impact analysis can enhance maintenance efficiency for modified functionalities during software evolution. Previous methods predominantly rely on static syntactic structures within function call relationships while neglecting semantic constraints' influences on actual execution paths, consequently failing to provide quantitative evaluations against real software runtime behaviors. To address this limitation, this paper proposes a method for software constraint extraction and model generation based on Large Language Models (LLMs), aiming to deliver fine-grained quantitative analysis. Specifically, this paper designs a software model for change impact analysis that performs code-to-model mapping through LLMs, extracting both variable propagations and constraints from function call relationships to obtain more comprehensive software logic. Furthermore, this paper employs a simulator that reconstructs software logic for simulation analysis integrated with Monte Carlo method, statistically measuring invocations of software elements. Finally, this paper proposes multiple metrics for systematic evaluation, enabling a quantitative change impact analysis. In case study, this paper validates the approach through typical function call relationships, the results demonstrate the effectiveness of the proposed approach.