Precise and Scalable Points-to Analysis via Data-Driven Context Tunneling

on YouTube ↗17m

" summary

Presents a concrete method called context tunneling that selectively preserves “important” context elements instead of the most-recent k, implemented in the Doop framework and learned automatically from code with a custom search over 23 syntactic features. Evaluation across four flavors (call-site, object, type, hybrid) shows 1-context outperforms k=2 in both precision and runtime on benchmark suites, and the learning algorithm handles the non‑monotonic precision landscape by alternating exploration and exploitation. The paper ships a working implementation, training protocol, and measurements that report sizable precision and scalability gains.

AI-written summary · about

" chapters (20)

1. 0:00Start
2. 0:36Static Analysis Needs Context Abstraction
3. 1:08Conventional Abstraction: Keep Most Recent K Contexts
4. 1:47Example
5. 3:371-CFA with Context Tunneling
6. 5:20Challenge: When to apply
7. 6:25Data-Driven Context Tunneling
8. 6:56Performance Highlight
9. 7:43Parametric Static Analyzer
10. 8:30Parametric Tunneling Heuristic
11. 9:04(1) Represent Method as Features
12. 9:42(2) Generate Tunneling Abstraction
13. 10:02Context Tunneling: Selectively Update Contexts
14. 10:24Learning Problem
15. 10:47Challenge: Non-monotonicity
16. 11:40Specialized Learning Algorithm
17. 13:12Effectiveness
18. 13:52Learned Heuristic
19. 14:41Conclusion
20. 15:39Method Features