The design and implementation of a relational programming system.

The declarative class of computer languages consists mainly of two paradigms
- the logic and the functional. Much research has been devoted in recent years
to the integration of the two with the aim of securing the advantages of both
without retaining their disadvantages. To date this research has, arguably, been
less fruitful than initially hoped. A large number of composite functional/logical
languages have been proposed but have generally been marred by the lack of a
firm, cohesive, mathematical basis. More recently new declarative paradigms,
equational and constraint languages, have been advocated. These however do
not fully encompass those features we perceive as being central to functional and
logic languages. The crucial functional features are higher-order definitions, static
polymorphic typing, applicative expressions and laziness. The crucial logic features are ability to reason about both functional and non-functional relationships
and to handle computations involving search.
This thesis advocates a new declarative paradigm which lies midway between
functional and logic languages - the so-called relational paradigm. In a relationallanguage program and data alike are denoted by relations. All expressions
are relations constructed from simpler expressions using operators which form
a relational algebra. The impetus for use of relations in a declarative language
comes from observations concerning their connection to functional and logic programming. Relations are mathematically more general than functions modelling
non-functional as well as functional relationships. They also form the basis of
many logic languages, for example, Prolog.
This thesis proposes a new relational language based entirely on binary relations, named Drusilla. We demonstrate the functional and logic aspects of
Drusilla. It retains the higher-order objects and polymorphism found in modern functional languages but handles non-determinism and models relationships
between objects in the manner of a logic language with notion of algorithm being composed of logic and control elements. Different programming styles -
functional, logic and relational- are illustrated.
However, such expressive power does not come for free; it has associated with
it a high cost of implementation. Two main techniques are used in the necessarily
complex language interpreter. A type inference system checks programs to ensure
they are meaningful and simultaneously performs automatic representation selection for relations. A symbolic manipulation system transforms programs to improve.
efficiency of expressions and to increase the number of possible representations
for relations while preserving program meaning.