Versioned E-Graphs

E-Graphs are an efficient encoding for discovering and maintaining sets of equalities, commonly adopted in the context of formal proofs, program analysis, and optimization. In several scenarios, equalities may hold only conditionally, i.e., under certain assumptions. For example, in automated provers the proof is often split into multiple branches, such that each branch considers a different set of equalities. A limitation of traditional e-graphs is that they can only encode a single set of equalities at a time. Conditional equalities are then handled by maintaining multiple e-graphs, e.g., one for each branch in the proof, which is inefficient as equalities shared among branches are simply replicated many times. In this paper, we introduce versioned e-graphs, which efficiently encode multiple equivalence sets at the same time, maximizing shared information among them. We formalize for versioned e-graphs and prove their correctness. We evaluate our solution against widely-adopted solutions which maintain multiple e-graphs and show that versioned e-graphs are up to 5 − 30 % more memory efficient and up to 4 × faster depending on the case study, especially when solution spaces are large both in explored program terms and number of branches.