Partial Evaluation for the Understanding of Fortran Programs

Abstract

This paper describes a technique and a tool that support partial evaluation of FORTRAN programs, i.e., their specialization for specific values of their input variables. The authors’ aim is to understand old programs, which have become very complex due to numerous extensions. From a given FORTRAN program and these values of its input variables, the tool provides a simplified program, which behaves like the initial program for the specific values. This tool mainly uses constant propagation and simplification of alternatives to one of their branches. The tool is specified in terms of inference rules and operates by induction on the FORTRAN abstract syntax. These rules are compiled into Prolog by the Centaur/FORTRAN programming environment. The completeness and soundness of these rules are proven using rule induction.

Fortran program specialization

“…The application of partial evaluation to software maintenance is discussed by Blazy and Facon [10]. Their partial evaluator aims at improving the readability of Fortran programs, but not at improving the efficiency of the programs.…”

Specialization of imperative programs through analysis of relational expressions

“…Blazy and Facon use constant propagation as a specialization enabler [4]. If a controlling expressions of a conditional branch reduces to an equality between a variable and a constant, then their method propagates the constant in the then-branch.…”

“…Baier, Glueck, and Zochling created a partial evaluator for Fortran [3]. Blazy and Facon created a Fortran specializer aimed at program understanding [4]. In contrast to these partial evaluation techniques for imperative languages, our specializations are done at the level of control flow graphs.…”

Program slicing: methods and applications

“…Quasi static slicing has been applied for program comprehension together with transformations [41]. From this point of view, this approach can be considered as an extension of work presented in [12] where partial evaluation is used to understand programs. Combining partial evaluation with program slicing allows to restrict the focus of the specialized program with respect to a subset of program variables and a program point.…”