Abstract

A novel scheme for representing unbound variables was proposed by Taylor: that aliased unbound variables are represented as circular chains of cells. Unfortunately, few details were given on its implementation and the involved trade offs. We compare the proposed scheme to that of Warren's abstract machine, and find that Taylor's scheme (a) leads to increased trailing, (b) makes stack allocation more difficult, (c) requires more expensive copying of heap cells at times, and (d) has a more expensive variable binding mechanism. On the positive side, a very strong advantage is that Taylor's scheme entirely dispenses with pointer chain dereferencing. We conclude by enumerating the data flow properties a global analyzer should find in order to eliminate these difficulties. 1 Introduction Taylor presented an intriguing representation of unbound logic variables in his thesis [5]. The standard WAM scheme [9] represents variables as pointers and unbound variables as self-pointers. Variable bi...

Citations