The two-level approach to data definition and space management in the LIS system implementation language

Abstract

It has often been remarked that surprisingly few parts of the algorithms which define an operating system require the use of unsafe, low-level language features {1}. Nevertheless, the lack of safety inherent in these parts will propagate to the whole system if an appropriate methodology is not used. It seems desirable to structure programs in such a way that parts with different safety levels are clearly separated and identifiable. The LIS solution to this problem involves the use of two language levels which are used in separate contexts. The general language is a Pascal-like language with the safety that this implies {2, 3}. On the other hand, the use of unsafe features is restricted to so-called implementation parts associated with each declaration level of the program. Several authors {4, 5, 6} have pointed out the advantages of having a two-step data definition, namely, a first step concerned with the semantic properties and a second dealing with the effective realization. In what follows we present the LIS approach to this question and show that the safety levels inherent to these two steps are different. The second step requires implementation specifications which appear in implementation parts.