In software systems, tools such as compilers, testers, and debuggers require static semantic information, both to insure their correctness and to enhance performance. Incremental update algorithms, which only calculate data flow information affected by the program changes, avoiding total recalculation, efficiently provide consistent documentation for a large evolving software system. Previously, incremental techniques for Fortran-like languages and exhaustive analyses for C programs have been developed. When these analyses of C systems produce solutions, they are inexpensive and high quality; however, often the analyses exhaust available resources without yielding a complete solution, especially for larger programs (over 5,000-10,000 lines of code). This research aims at expanding the applicability of exhaustive techniques to handle a wider class of programs and developing incremental techniques for C systems. The work is performed by an academic and industrial team, focusing on the development of new analysis techniques and their testing on research industrial applications. Goals are to: (i) scale the existing prototype to handle programs with 10,000 to 100,000 lines of code, (ii) develop techniques for analyzing modules without analyzing the entire program (i.e., a separate compilation mode of analysis),(iii) develop incremental techniques for interesting interprocedural static analyses for languages with general purpose pointers, and (iv) enhance the theoretical infrastructure for semantic analysis of modern programming languages (i.e., C, C++, and Fortran 90). The research explores techniques for varying the precision of alias information by designing new, flow-insensitive methods, to augment our current flow-sensitive ones. Partitioning the program variables allows for an aliasing solution with varying degrees of precision for different sets of variables. Experiments are conducted varying the value of k for these different sets, using k-limiting to handle dereferencing in recursive structures. A hybrid incremental analysis algorithm is used as a basis for incrementalizing side effect analysis of C programs.
