ARTFEED — Contemporary Art Intelligence

CAT: Call-Chain-Aware LLM Test Generation for Java

other · 2026-04-27

A new approach called CAT uses call-chain and dependency contexts to improve LLM-based unit test generation for Java projects. Existing methods rely on execution-path information, which is insufficient for complex systems with inter-class dependencies. CAT models caller-callee relationships, object constructors, and third-party dependencies through static analysis, and supports iterative test fixing. It was evaluated on Defects4J and four real-world GitHub projects released after the LLM's cutoff date.

Key facts

  • CAT is a call-chain-aware LLM-based test generation approach
  • It incorporates call-chain and dependency contexts via static analysis
  • It models caller-callee relationships, object constructors, and third-party dependencies
  • It supports iterative test fixing
  • Evaluated on Defects4J benchmark
  • Evaluated on four real-world GitHub projects
  • Projects released after LLM's cutoff date
  • Addresses limitations of execution-path-based approaches

Entities

Sources