Rewriting Declarative Query Languages
Matthias Brantner · MADOC (University of Mannheim) · 2007
Queries against databases are formulated in declarative languages. Examples are the relational query language SQL and XPath or XQuery for querying data stored in XML. Using a declarative query language, the querist does not need to know about or decide on anything about the actual strategy a system uses to answer the query. Instead, the system can freely choose among the algorithms it employs to answer a query. Predominantly, query processing in the relational context is accomplished using a relational algebra. To this end, the query is translated into a logical algebra. The algebra consists of logical operators which facilitate the application of various optimization techniques. For example, logical algebra expressions can be rewritten in order to yield more efficient expressions. In order to query XML data, XPath and XQuery have been developed. Both are declarative query languages and, hence, can benefit from powerful optimizations. For instance, they could be evaluated using an algebraic framework. However, in general, the existing approaches are not directly utilizable for XML query processing. This thesis has two goals. The first goal is to overcome the above-mentioned misfits of XML query processing, making it ready for industrial-strength settings. Specifically, we develop an algebraic framework that is designed for the efficient evaluation of XPath and XQuery. To this end, we define an order-aware logical algebra and a translation of XPath into this algebra. Furthermore, based on the resulting algebraic expressions, we present rewrites in order to speed up the execution of such queries. The second goal is to investigate rewriting techniques in the relational context. To this end, we present rewrites based on algebraic equivalences that unnest nested SQL queries with disjunctions. Specifically, we present equivalences for unnesting algebraic expressions with bypass operators to handle disjunctive linking and correlation. Our approach can be applied to quantified table subqueries as well as scalar subqueries. For all our results, we present experiments that demonstrate the effectiveness of the developed approaches.