Reproduce “PDLogger: Automated Logging Framework for Practical Software Development” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shengcheng Duan, Yihua Xu, Sheng Zhang, Shen Wang, Yue Duan

PDLogger introduces an automated logging framework for practical software development, explicitly addressing multi-log scenarios within methods, a limitation of prior work. The framework integrates precise semantic dependency analysis via program slicing and a function-aware variable extension, leading to a 69.21% higher F1 score for log position and a 135% higher F1 for variables in multi-log generation compared to adapted baselines.

Key insights

  • Structured prompts based on code block types are crucial for effectively guiding LLMs to predict multiple log positions within a method, significantly reducing false positives.
  • Integrating precise, statement-level semantic dependencies (via program slicing) into the LLM context dramatically improves the relevance and diagnostic utility of generated log messages and variable suggestions.
  • Expanding the scope of candidate logging variables to include non-member functions and function expressions, beyond local and member variables, is essential for generating diagnostically comprehensive log statements.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.