We investigate a new query reverse-engineering problem of unmasking SQL queries hidden within database applications. The diverse use-cases for this problem range from resurrecting legacy code to query rewriting. As a first step in addressing the unmasking challenge, we present UNMASQUE, an active-learning extraction algorithm that can expose a basal class of hidden warehouse queries. A special feature of our design is that the extraction is non-invasive wrt the application, examining only the results obtained from repeated executions on databases derived with a combination of data mutation and data generation techniques. Further, potent optimizations are incorporated to minimize the extraction overheads. A detailed evaluation over applications hosting hidden SQL queries, or their imperative versions, demonstrates that UNMASQUE correctly and efficiently extracts these queries.

Given a database instance and a populated result, query reverse-engineering attempts to identify candidate SQL queries that produce this result on the instance. A variant of this problem arises when a ground-truth is additionally available, but hidden within an opaque database application. In this demo, we present UN-MASQUE, an extraction algorithm that is capable of precisely identifying a substantive class of such hidden queries. A hallmark of its design is that the extraction is completely non-invasive to the application. Specifically, it only examines the results obtained from application executions on databases derived with a combination of data mutation and data generation techniques, thereby achieving platform-independence. Further, potent optimizations, such as database size reduction to a few rows, are incorporated to minimize the extraction overheads. The demo showcases these features on both declarative and imperative applications.