Metaprogramming, defined as creating application programs by writing programs that produce programs, is presented as the basis of a method for reducing software costs and improving software quality. The method starts with a rapid prototyping phase in which selected representative parts of the application are prototyped; this is followed by a tooling up phase, during which the metaprogramming occurs. The final phase is the production of the application programs using the software tools and techniques of the metaprogramming phase. The author summarizes the experience of two projects that support the assertion that metaprogramming is an efficient way to produce programs, since these projects used metaprogramming in their development. An economic theory which justifies this approach is outlined. It is shown how to do a cost/benefit analysis of the method.

No abstract available.

The method of local constraints attempts to describe context-free languages in an apparently context-sensitive form which helps to retain the intuitive insights about the grammatical structure. This form of description, while apparently context-sensitive is, in fact, context-free and allows a program derivation structure to be represented as a tree with additional constraints, thus allowing for the possibility of a correctness proof in the form of Knuthian semantics. A part of ALGOL 60 syntax has been restated as a grammar with local constraints. It is also shown that parsing of grammars with local constraints can be carried out by a suitable extension of Early's algorithm. Finally, the relationship of local constraints to the semantics of programs is discussed in Section 4 by considering some program transformations.

A cross compiler producing programs for the HP-65 from source programs in BASIC is described. Optimisation techniques are discussed in detail, and typical programs presented showing the resultant code. Typical compilation costs are cited.

Game playing and backtracking programs among others can be understood in terms of algebraic operations with zeros. The special properties of the algebraic operations can be used to explain the equivalence of otherwise dissimilar algorithms.

It is very often more convenient and more meaningful to specify a set of structural descriptions analytically rather than generatively, i.e., by specifying a set of constraints each structured description in the set has to satisfy. Peters and Ritchie [7] have shown that if context-sensitive rules are used only for “analysis” then the string language of the set of trees is still context-free. In this paper, we have generalized this result by considering context-free rules constrained by Boolean combinations of proper analysis predicates and domination predicates. These rules, called "local transformations" not only make precise an informal and briefly discussed notion of Chomsky [2], but also, generalize it in an appropriate manner. It is shown that the Peters-Ritchie result can be generalized to local transformations. Linguistic relevance of this result has been also briefly discussed. Results in this paper are relevant to the following situation: Patterns of a class, say A, may be difficult to characterize generatively; but, it may be possible to specify a suitable (nontrivial) augmentation, say B, of the class A, such that B can be easily characterized generatively, and then A is characterized by stating some restrictions on the class B. This suggests possible applications to programming languages and pattern description languages.

The class of regular sets is shown to be contained in the class of local adjunct languages. The essential idea in the proof is an intercalation theorem for the Myhill congruence classes.

A theoretical model of a microprogram unit is formulated, and the relationship of this model to a form of feedback shift register is developed. The feedback shift-register machine variant has synchronous and asynchronous prototypes. The asynchronous binary feedback shift-register machines realize every fundamental mode asynchronous sequential (finite-state) machine. The synchronous feedback shift-register machine differs from the standard definition by the addition of transitional states to the usual (nontransitional) states, and a single step in the simulation is a move between nontransitional states. When synchronous feedback shift-register machines are generalized in this way, they also become universal components capable of simulating ali finite-state machines.

In this paper, we will essentially present three results in tree automata. Two of these are concerned with certain open problems posed by Rounds [1] and Martin and Vere [3], and the third one is related to the result of Peters and Ritchie [4].