Adapting CLP(R) To Floating-Point Arithmetic

As a logic programming language, Prolog has shortcomings. One of the most serious of these is in arithmetic. CLP(R), though a vast improvement, assumes perfect arithmetic on reals, an unrealistic requirement for computers, where there is strong pressure to use floatingpoint arithmetic. We present an adaptation of CLP(R) where the errors due to floating-point computation are absorbed by the use of intervals in such a way that the logical status of answers is not jeopardized. This system is based on Cleary's "squeezing" of floating-point intervals, modified to fit into Mackworth's general framework of the Constraint-Satisfaction Problem. Our partial implementation consists of a meta-interpreter executed by an existing CLP(R) system. All that stands in the way of correct answers involving real numbers is the planned addition of outward rounding to the current prototype. 1 Introduction Mainstream computing holds that programming should be improved by gradual steps, as exemplified by the m..