Generalizable synthesis through unification

The generalizability of PBE solvers is the key to empirical synthesis performance. Despite importance generalizability, related studies on are still limited. In theory, few existing provide theoretical guarantees and in practice, there a lack with satisfactory important domains such as conditional linear integer arithmetic (CLIA). this paper, we adopt concept from computational learning Occam learning, perform comprehensive study framework through unification (STUN), state-of-the-art for synthesizing programs nested if-then-else operators. We prove that Eusolver, STUN solver, does not satisfy condition then design novel PolyGen, which theoretically guaranteed by learning. evaluate PolyGen CLIA demonstrate significantly outperforms two CLIA, Eusolver Euphony, both efficiency.