CoDocBench : A Dataset for Code-Documentation Alignment in Software Maintenance

Kunal Pai, Premkumar Devanbu, and Toufique Ahmed University of California, Davis, USA · 2025

Read CoDocBench : A Dataset for Code-Documentation Alignment in Software Maintenance by Kunal Pai, Premkumar Devanbu, and Toufique Ahmed University of California, Davis, USA (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-00519.txt.

Description

Pai, Devanbu, and Ahmed build in twenty twenty-five a dataset that pairs code changes with matching changes to code comments. Keeping code and its written notes in sync is core to software upkeep. Given a note about a wanted change, a helper should draft the code edits. Given edits, it should draft the new note.

Their dataset mines real, high-quality public projects for commits where code and its note changed together. Each sample is one such paired change, drawn from real work rather than made up. They set out how the data were gathered and pose hard, real tasks for training and testing. Two large current models still found those tasks hard, which shows the dataset's value.

Catalog record

CREATORKunal Pai, Premkumar Devanbu, and Toufique Ahmed University of California, Davis, USA
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-00519

Browse the full library · About the Tower of Babel