FTFY: Research advances in automatic bug repair
summary
Research presents concrete methods and results for automatic source‑level bug repair using AST‑level genetic programming and combinations of program synthesis, machine learning, and search over large code corpora. It treats failing tests as specifications, claims most real patches are 3–5 lines and code is repetitive so reuse is common — roughly five to seven lines before new textual uniqueness — and reports tools like GenProg that actually find working patches for multiple real programs, not just toy examples.
AI-written summary · about
chapters (6)
- 0:00Start
- 0:13O'REILLY Velocity
- 1:43Problem: source-level defect repair
- 2:31Insight 1: Code is repetitive + predictable
- 3:00Secret Sauces
- 3:52GenProg: automatic program repair using genetic programming
related talks
12mAutomatic Grammar Repair
18mPreview: Verification (Correctness & Debugging)
59mAutomatic Grammar Repair
20mAutomatic Repair of Regular Expressions
44mTest Intelligence: Finding More Bugs in Less Time with AI
52mStructure aware fuzzing
40mFrom Bugs to Brilliance: Leveraging AI in Modern Software Testing
41mA Human-Centric Quality Assurance Process for Open Source Softwa...
