Amortised memory analysis using the depth of data structures

Abstract

Hofmann and Jost have presented a heap space analysis [1]that finds linear space bounds for many functional programs. It uses an amortised analysis: assigning hypothetical amounts of free space (calledpotential) to data structures in proportion to their sizes using type annotations. Constraints on these annotations in the type system ensure that the total potential assigned to the input is an upper bound on the total memory required to satisfy all allocations. We describe a related system for bounding the stack space requirements which uses the depth of data structures, by expressing potential in terms of maxima as well as sums. This is achieved by adding extra structure to typing contexts (inspired by O'Hearn's bunched typing [2])to describe the form of the bounds. We will also present the extra steps that must be taken to construct a typing during the analysis.