Ask Me Anything with Peter O'Hearn
summary
Answers claim a shift from heavy upfront specs to a pay-as-you-go model where small specs give incremental value, and point to concrete tools: Infer team work at Facebook scales compositional static analysis to ~10M lines, RacerD catches concurrency bugs and RacerD found 2.5k bugs on Android, batch reports had ~0% fix rate versus ~70% when integrated into diffs, under-approximation theory complements over-approximation, buffer overruns remain hard at scale, and fuzzing and static analyses prove complementary.
AI-written summary · about
chapters (15)
- 0:00Introduction to Peter O'Hearn
- 1:33Technical Q&A: Part one
- 6:20Approximation in analysis
- 9:05Academia vs. industry
- 11:58Industry insights
- 15:36Blurring types and verification
- 17:09Formal methods effectiveness
- 19:18Static analysis advancements
- 22:59Memory and concurrency
- 25:02Origins of separation logic
- 27:05Static versus dynamic analysis
- 30:36Collaboration and career advice
- 35:09Hybrid fuzzing and analysis
- 39:48Refining analysis tools
- 43:43Reflections and personal life
related talks
29mEvent-based Fuzzing, Patch-based Research, and Comment Police: Finding Bugs Through a Bug
40mCatching More Bugs with Fewer False Alarms
01h 03mSOAP - Keynote: Beyond Code: New Signals for Static Analysis
52mStructure aware fuzzing
58mSecurity Static Analysis - Avoiding an Angry Mob of Engineers
42mContemporary Automatic Program Analysis
22mCombining Agentic AI and Lightweight Formal Methods To Find Bugs in a Production(…)- 40mCatching More Bugs with Fewer False Alarms