Variadic genericity through linguistic reflection: a performance evaluation

The use of variadic genericity within schema definitions increases the variety of databases that may be captured by a single specification. For example, a class of databases of engineering part objects, in which each database instance varies in the types of the parts and the number of part types, should lend itself to a single definition. However, precise specification of such a schema is beyond the capability of polymorphic type systems and schema definition languages. It is possible to capture such generality by introducing a level of interpretation, in which the variation in types and in the number of fields is encoded in a general data structure. Queries that interpret the encoded information can be written against this general data structure.
An alternative approach to supporting such variadic genericity is to generate a precise database containing tailored data structures and queries for each different instance of the virtual schema.1 This involves source code generation and dynamic compilation, a process known as linguistic reflection. The motivation is that once generated, the specific queries may execute more efficiently than their generic counter-parts, since the generic code is “compiled away”. This paper compares the two approaches and gives performance measurements for an example using the persistent languages Napier88 and PJama.