Diff-Based Code Editing Underperforms Direct Generation Except on Localized Edits
Andrej Andrejev · hf · 2026-09-10
An empirical study comparing iterative diff-based editing vs. direct whole-file generation for Flutter/Dart code models finds that diff-based editing underperforms overall.
It only excels on short, localized edits such as refactoring and error fixes — a property the authors term 'task locality,' with implications for how code models should be trained and deployed.
More from coding & agent
- Hermes Agent routing flaw silently hijacks models to metered OpenRouter, costing users money — WolframRvnwlf · 2026-09-10
- mcp-oracle-h adds a mandatory human approval gate for irreversible agent actions via Telegram — modelcontextprotocol · 2026-09-10
- Codex already lets you group pinned sessions by dragging them into projects — vwxyzjn · 2026-09-10
- Use /goal to steer the astra coding agent back on track — TheMoonMidas · 2026-09-10
- How a Buffer Overflow Overwrites the Return Address in C and x86-64 — tetsuoai · 2026-09-10
- Full pipeline: niji + GPT Image + local MiniMax H3 + Codex for pixel-perfect animation — Hailuo_AI · 2026-09-10