Труды Института системного программирования РАН (Oct 2018)
Features of SQL table expressions and their compliance with the concepts of relational model
Abstract
The paper analyses a behavior of the SQL standard’s table expressions in comparison with relational equivalents. The concept of relation of relational model presumes nonexistence of duplicate tuples, attributes of the same name, ordering of attributes, ordering of tuples, no-typed NULLs. The paper demonstrate cases where behavior of constructions of the SQL-standard violates the rules of the relational model: these constructions produce tables that are not relations (join expressions, table references, set-theoretical operators, simple tables, and select). The paper also describes practices that allow to avoid such cases and to use SQL in truly relational manner. Relational use of SQL is possible nearly always. However, existing implementations of SQL are far from ideal, and sometimes such use leads a poor performance. Non-relational use of SQL means a tradeoff between performance and conceptual consistence. It is always useful to understand a theoretically perfect variant of use and to have solid reasons to abandon it. It is also useful to document these reasons to return to the perfect variant with some new version of SQL-based DBMS that does not demonstrates problems of performance.
