Synthesizing Code Quality Rules from Examples
summary
Automatic synthesis of code-quality rules from code-change examples produces executable graph queries now deployed in Amazon CodeGuru Reviewer. The system uses RhoSynth to learn existential first-order patterns over program-dependence graphs via Integer Linear Programming-based graph alignment, bootstraps positive/negative examples from developer edits, synthesizes and refines rules, generates 30+ Java rules sent to production, and yields detections with around 75% precision from live developer feedback.
AI-written summary · about
chapters (15)
- 0:00Intro
- 0:19Motivation
- 2:17Problem: general, practical, natural
- 3:12An example
- 5:25Existing approaches
- 7:24Synthesis problem
- 13:32check-movetofirst Rule
- 20:35Maximizing graph alignment using ILP
- 23:03RhoSynth algorithm: characteristics
- 23:59Rule synthesis setting
- 26:00Rule deployment
- 26:56Result highlights
- 28:09Results: rule characteristics
- 28:20In the paper, compare against...
- 28:30Summary
related talks
51mMake Quality Executable: Verification Gates for Agentic Engineering at Scale
46mCode Quality in the AI era, needed more than ever
47mWhy Bad Code Quality Is Not (Only) Your Fault
38mCracking the Code Review: from Guesswork to Automated Guidance
31mMay the Code Review Be with You
19mWith great power comes great responsibility in code quality
43mCracking the Code Review: from Guesswork to Automated Guidance
57mVariant analysis