SQL-AG: Querying structured documents using attribute grammars

Structured documents, such as program source texts, technical documentation, or XML data, comprise an important class of data in many applications. Structured documents are distinguished from flat text by their tree structure. In a program source text, this structure is the abstract syntax tree of the program. In a technical document, this structure is the division in chapters, sections, paragraphs, definitions, examples, footnotes, and so on. In an XML document, this structure is the tree structure provided by the tag markup. End-users may be satisfied with simple keyword-type searches over structured documents. Application developers, however, need to be able to perform more powerful types of searches and transformations of structured documents, where the structure is explicitly taken into account. For example, in a Java source file, we may want to look up all method names. In a technical document, we may want to move all footnotes to an appendix. On XML data, structural queries and transformations are the bread and butter of special-purpose XML manipulation languages such as XSLT or XQuery. Indeed, XSLT and XQuery are existing valuable tools for expressing structural querying of structured documents. Nevertheless, there are interesting alternatives which merit to be explored. Specifically, one such alternative is our SQL-AG system. In a nutshell, in SQL-AG, the tree structure is defined by a context-free grammar. A structured document then is viewed as a parse tree according to this context-free grammar. The document is stored in a relational database. Structured queries are now expressed by annotating the context-free grammar with attribute definitions: this is the standard attribute grammar mechanism. In SQL-AG, attribute values are relations over the nodes and text values of the document. Attribute definitions are expressed in standard SQL. SQL-AG is not meant as an ad-hoc query language, but rather as a powerful tool for application developers.