Automatic generation of test data for relational queries (Q1823740)

The problem of generating good test databases is presented. The concept is characterized by Armstrong databases generated by using two techniques. The first technique, based on taking unions of databases, is very general, but produces rather large databases. The second technique is based on chasing a collection of counterexamples; it produces smaller databases and applies rather well to some types of queries. These methods of producing test databases are based on the semantics of the queries and not on their structure.