Worst-case execution time test generation using genetic algorithms with automated construction and online selection of objectives

Worst-case execution time test generation can be difficult if tested programs use complex heuristics. Such programs may fail only on very small subsets of possible input data. Previous works show that evolutionary optimization (in particular, genetic algorithms) is a suitable tool for test generation under such conditions. We present an approach of automated integration of counters in the source code. There are two types of counters: one for counting the number of procedure calls, and another one for counting the number of loop executions. The values of these counters at the end of the program execution, as well as the execution time, serve as optimization objectives. We also propose two new methods for online selection of objectives. Together with the counter integration approach, they augment the already existing test generation method and increase its degree of automation. The experimental results for three example programs and for several objective selection algorithms are presented.