A proof method for the correctness of modularized 0CFA

Modular program analysis, which analyzes separated program sources such as modules, is a practical alternative to whole-program analysis. It does not need the entire program text as its input, and if some parts of the program are modified, it re-analyzes only the dependent parts of a modified module. This article is about our findings when we tried to derive a modular version from a whole-program control-flow analysis (CFA) [1–3], to be used inside a modularized version of our exception analysis [4–6]: • Deriving a modular version from a whole-program monovariant (or context-insensitive) CFA makes the resulting analysis polyvariant (or contextsensitive) at the module level. • Hence the correctness of its modularized version cannot be proven in general with respect to the original CFA. • A convenient stepping stone to prove the correctness of a modularized version (instead of proving it