An Incremental Slicing Method for Functional Programs

TL;DR

This paper introduces an incremental static slicing method for functional programs that reuses previous analysis results to efficiently compute slices for new criteria, significantly improving performance.

Contribution

It presents the first incremental slicing algorithm for functional programs, utilizing function summaries and a pre-computation step for faster slicing.

Findings

Incremental slicing runs orders of magnitude faster than non-incremental methods.

The approach is validated on a Scheme subset with correct results.

Pre-computation enables low-cost updates for new slicing criteria.

Abstract

Several applications of slicing require a program to be sliced with respect to more than one slicing criterion. Program specialization, parallelization and cohesion measurement are examples of such applications. These applications can benefit from an incremental static slicing method in which a significant extent of the computations for slicing with respect to one criterion could be reused for another. In this paper, we consider the problem of incremental slicing of functional programs. We first present a non-incremental version of the slicing algorithm which does a polyvariant analysis 1 of functions. Since polyvariant analyses tend to be costly, we compute a compact context-independent summary of each function and then use this summary at the call sites of the function. The construction of the function summary is non-trivial and helps in the development of the incremental version. The incremental method, on the other hand, consists of a one-time pre-computation step that uses the non-incremental version to slice the program with respect to a fixed default slicing criterion and processes the results further to a canonical form. Presented with an actual slicing criterion, the incremental step involves a low-cost computation that uses the results of the pre-computation to obtain the slice. We have implemented a prototype of the slicer for a pure subset of Scheme, with pairs and lists as the only algebraic data types. Our experiments show that the incremental step of the slicer runs orders of magnitude faster than the non-incremental version. We have also proved the correctness of our incremental algorithm with respect to the non-incremental version.