SQL declaratively specifies what the desired output of a query is. This work shows that a non-standard interpretation of the SQL semantics can, instead, disclose where a piece of the output originated in the input and why that piece found its way into the result. We derive such data provenance for very rich SQL dialects---including recursion, windowed aggregates, and user-defined functions---at the fine-grained level of individual table cells. The approach is non-invasive and implemented as a compositional source-level SQL rewrite: an input SQL query is transformed into its own interpreter that wields data dependencies instead of regular values. We deliberately design this transformation to preserve the shape of both data and query, which allows provenance derivation to scale to complex queries without overwhelming the underlying database system.

We demonstrate a new incarnation of Habitat, an observational debugger for SQL. In observational debugging, users highlight parts of a presumably faulty query to observe the evaluation of SQL subexpressions and learn about the query's actual runtime behavior. The present version of Habitat has been redesigned from scratch and employs a query instrumentation technique that exclusively relies on the SQL facilities of the underlying RDBMS. We particularly shed light on new features like (1) the debugging of recursive SQL queries and (2) the observation of row groups (before and after aggregation). Habitat can turn any reasonably modern SQL:1999 RDBMS into its own language-level SQL debugger.