Context-free language reachability (CFL-reachability) is a fundamental formulation for program analysis with many applications. CFL-reachability analysis is computationally expensive, with a slightly subcubic time complexity concerning the number of nodes in the input graph.

This paper proposes staged solving: a new perspective on solving CFL-reachability. Our key observation is that the context-free grammar (CFG) of a CFL-based program analysis can be decomposed into (1) a smaller context-free grammar, $L$, for matching parentheses, such as procedure calls/returns, field stores/loads, and (2) a regular grammar, $R$, capturing control/data flows. Instead of solving these two parts monolithically (as in standard algorithms), staged solving solves $L$-reachability and $R$-reachability in two distinct stages. In practice, $L$-reachability, though still context-free, involves only a small subset of edges, while $R$-reachability can be computed efficiently with close to quadratic complexity relative to the node size of the input graph. We implement our staged CFL-reachability solver, STG, and evaluate it using two clients: context-sensitive value-flow analysis and field-sensitive alias analysis. The empirical results demonstrate that STG achieves speedups of 861.59x and 4.1x for value-flow analysis and alias analysis on average, respectively, over the standard subcubic algorithm. Moreover, we also showcase that staged solving can help to significantly improve the performance of two state-of-the-art solvers, POCR and PEARL, by 74.82x (1.78x) and 37.66x (1.7x) for value-flow (alias) analysis, respectively.