Constructing a Complex for Automated Debugging of Fragmented Programs

Abstract

The paper is devoted to the authors’ tools for automated debugging of fragmented programs in the LuNA (Language for Numerical Algorithms) system. The LuNA system uses approach of fragmented programming, which allows the researcher to design parallel programs in an automated mode.The main problem discussed in the paper is the detection of errors specific to LuNA programs. The operation algorithms of four developed tools for automated debugging of LuNA programs are described. Three of them are based on the static analysis approach and differ in the intermediate representations used. The first uses an abstract syntax tree (AST). The second uses a data dependency graph (DDG). The third tool generates a Prolog program containing a set of facts about the original LuNA program and rules corresponding to erroneous situations. Due to the use of various intermediate representations, each of the static analysis tools is able to detect its own types of errors. The last tool uses the “post-mortem analysis” approach. Each process of a parallel program collects a trace file during its operation and at the end of the work these files are analyzed by a special utility.The paper also describes the developed utility for testing automated debugging tools. The set of test programs with various errors, as well as error-free programs, has been developed  for this utility.The last part describes the designed mechanism for integrating the developed automated debugging tools into a single software package.