Towards Automatically Pruning Logging Code with Coding Agents: How Far Are We?
Coding agents usually produced valid edits, but rarely matched the real logging removals.
The paper studies logging-code removal as its own maintenance task, using manually validated Python and Java cases. Its LogRem dataset covers 387 real-world changes, including direct removals, infrastructure removals, and replacements. Across four coding agents, 95.6% to 100.0% of outputs passed validity checks, but only 11.1% to 19.6% removed the same logging code while preserving unrelated code. Commit messages and developer discussions helped alignment most; taxonomy guidance cut runtime. ArXiv · AI/CL/LG's note
The paper studies logging-code removal as its own maintenance task, using manually validated Python and Java cases. Its LogRem dataset covers 387 real-world changes, including direct removals, infrastructure removals, and replacements. Across four coding agents, 95.6% to 100.0% of outputs passed validity checks, but only 11.1% to 19.6% removed the same logging code while preserving unrelated code. Commit messages and developer discussions helped alignment most; taxonomy guidance cut runtime. ArXiv · AI/CL/LG's note
score 3