Search papers, labs, and topics across Lattice.
This study investigates deletion avoidance in large language models (LLMs) during code editing, revealing that these models often fail to remove necessary code, leading to less maintainable codebases. The research shows that while models can identify the correct files for deletions over 92% of the time, they only successfully delete the required lines in under 52% of cases, frequently opting for a "Guard-and-Go" strategy instead. By introducing the CanItDelete benchmark and demonstrating that targeted post-training on deletion can mitigate this issue, the authors highlight a critical gap in LLM training that impacts real-world code repair effectiveness.
LLMs struggle with code deletion, often opting for workarounds that compromise code maintainability, revealing a significant training gap in their editing capabilities.
Large language models increasingly write and repair production code, yet evidence is mounting that their test-passing patches leave codebases harder to maintain. We identify one concrete source: deletion avoidance, the systematic tendency to retain code that an intended edit requires removing. Across the five leading models on the official SWE-bench Verified leaderboard, deletion recall against the developer patch reaches at most 71.7% even on tasks all five solve, and models reach the right file for over 92% of required deletions but cut the exact line in under 52% of cases. Instead, 29.0% of passing patches wrap the targeted code in a guard or fallback, a pattern we call Guard-and-Go. Such patches pass because the original tests rarely check removal: when we retrofit 34 Verified tasks with tests that fail if the targeted code remains, four frontier models spanning closed and open weights fall from 63.2% to 41.9%. Because real repairs mix removal with addition, we curate CanItDelete, a benchmark of 200 tasks mined from real commits whose entire required edit is deletion. Even with the addition work gone, the best model still fails one task in five, and smaller open models fall to 18.0%. We then ablate GPT-5.6 Sol under four cumulative prompts; success moves little until we supply the exact lines, which nearly eliminate incomplete deletion yet raise success only to 80.5% because the model then deletes beyond the spans or adds code instead. Finally, through a pilot study we show one potential fix: teaching deletion during post-training reduces deletion avoidance and improves broader code-editing performance, suggesting the behavior is undertrained rather than beyond reach.
