Abstract:Abstract interpretation offers a principled foundation for static analysis by approximating concrete program semantics via abstract domains. However, quantitatively comparing the precision of different domains remains a longstanding challenge. We present MCAI (Model Counting meets Abstract Interpretation), a new methodology that employs model counting to measure the precision of abstract domains. Unlike prior approaches that assess precision relative to specific analysis queries, MCAI encodes both concrete semantics and abstract values as logical formulas, enabling a client-independent, quantitative metric of imprecision that captures the inherent semantic loss in the abstractions. We apply MCAI to four abstract domains and evaluate the precision of their best abstract transformers via symbolic abstraction. Our results yield several insights: the Interval domain, despite its simplicity, often achieves precision comparable to that of Octagon; many octagonal constraints are semantically redundant; and the bit-level KnownBit domains consistently outperform the word-level domains. MCAI offers both theoretical insights into the precision of abstract domains and practical guidance for selecting suitable abstractions.
From: Peisen Yao [view email]
[v1]
Sat, 20 Jun 2026 11:08:50 UTC (2,197 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。