Effective concolic unit testing with dynamic unit contexts carved from system tests

Abstract

Automated unit testing techniques like concolic unit testing improve the benefits of unit testing through diverse unit test executions. However, current automated unit testing techniques do not utilize valuable information on dynamic unit contexts (DUCs) (i.e., values of all parameters and global variables read by a target function) in system tests due to the technical difficulty to extract them from system tests.

I have developed a new concolic unit testing framework $CUT^2$ for complex C programs. First, $CUT^2$ instruments a target program P to insert probes that capture/carve DUCs of f for system tests. Second, $CUT^2$ carves DUCs of f while executing the instrumented target program with system tests. At this step, to carve DUCs of f accurately, $CUT^2$ carves not only parameters of f but also global variables updated by f and f's descendant functions. Third, $CUT^2$ generates a symbolic test driver and stubs that build symbolic search space based on the carved DUCs of f and performs concolic unit testing on f using the carved DUCs as initial test inputs for f.

In the experiments on CoREBench, $CUT^2$ achieves around 90% branch coverage on average, which is at lease 10.9%p higher than the existing concolic unit testing techniques on average.