Automating relational operations on data structures

An approach to having compilers do most of the implementation detail work in programming that divides a program into two parts is described. The specification part describes what the program should do, but in a way that avoids commitment to implementation details. The annotation part provides implementation instructions that the compiler will carry out. Annotations affect execution efficiency, but not functional behavior. They are very high level and usually very short and hence encourage experimentation. To try out different implementation choices, the programmer simply changes the annotations and recompiles. The implementation details related to data representations are discussed. The testing of compilers that produce Lisp code for iteration, and for adding and deleting tuples of composite relations, is reviewed.<>