Gistify! Codebase-Level Understanding via Runtime Execution

TL;DR

Gistify is a new task that challenges coding language models to generate minimal, self-contained files that replicate specific functionalities of large codebases by understanding their structure and execution flow.

Contribution

The paper introduces Gistify, a novel code understanding task that evaluates models' ability to produce minimal reproductions of codebase functionalities.

Findings

State-of-the-art models struggle with Gistify tasks.

Performance drops on tasks with long execution traces.

Gistify highlights limitations in current code understanding capabilities.

Abstract

As coding agents are increasingly deployed in large codebases, the need to automatically design challenging, codebase-level evaluation is central. We propose Gistify, a task where a coding LLM must create a single, minimal, self-contained file that can reproduce a specific functionality of a codebase. The coding LLM is given full access to a codebase along with a specific entrypoint (e.g., a python command), and the generated file must replicate the output of the same command ran under the full codebase, while containing only the essential components necessary to execute the provided command. Success on Gistify requires both structural understanding of the codebase, accurate modeling of its execution flow as well as the ability to produce potentially large code patches. Our findings show that current state-of-the-art models struggle to reliably solve Gistify tasks, especially ones with long executions traces.