Incremental Compositional Dynamic Test Generation

Dynamic test generation consists of running a programwhile simultaneously executing the program symbolically in order to gather constrains on inputs from conditional statements encountered along the execution. Those constraints are then systematically negated and solved with a constraint solver, generating new test inputs to exercise different execution paths of the program. This process is repeated with the goal of exercizing many program paths and finding many bugs. This search process can be made compositional by memoizing symbolic execution sub-paths as test summaries that are re-usable during the search, resulting in a search algorithm that can be exponentially faster than a non-compositional one. By construction, symbolic test summaries are “must” summaries guaranteeing the existence of some program executions. In this paper, we consider the problem of statically validating symbolic test summaries against code changes. Summaries that are provably still valid using a static analysis of a new program version do not need to be retested or recomputed dynamically. In the presence of small code changes, incrementality can considerably speed-up regression testing since static checking is much cheaper than dynamic checking and testing. We propose a two-phase algorithm to statically check which old test summaries are still valid for a new program version. The first phase is based on a simple analysis of the static call-flow and control-flow graph of the program. The second phase checks remaining invalidated intraprocedural test summaries against new code using verificationcondition generation and automated theorem proving. We discuss challenges in validating must summaries statically and provide several checks with different guarantees. We present preliminary experimental results with three large Windows applications.