Derivations with Holes for Concept-Based Program Synthesis

Abstract

Program synthesis has the potential to democratize programming by enabling non-programmers to write software. But conventional approaches to synthesis may fail if given insufficient information - a common occurrence when asking non-experts to describe the application they want to write. This paper introduces a new concept-based program synthesis mechanism that can cope with incomplete knowledge, targeting low-code model-driven languages. Concepts are modelled in an ontology that represents user intent including basic actions (e.g. show, filter, and create) along with their associated data as well as basic user interface structures like screens or pages. Our synthesis framework consists of a system of derivation rules that supports deferred premises, which need not be immediately satisfied during synthesis. A derivation in which some deferred premises are missing will thus contain holes; semantically, it represents a proof that is conditional on the developer filling the holes with additional facts from the ontology. We translate derivations with holes to standard first-order logic derivations, where the holes are transformed into assumptions. We illustrate the feasibility and effectiveness of our framework with a proof-of-concept implementation and a set of illustrative examples.