Analysis of runtime data-log for software fault localization

Software can contain faults that remain undetected prior to its release. It is then important to determine the plausible root-cause of fault, namely, the faulty lines of code, or indicators for any missing lines of code. To localize a software fault to its "root-cause", we introduce the notion of a fault-seed, a fragment of a faulty-run, and propose a model based automated approach that analyzes the observed faulty run of the software, recorded during its runtime operation, to determine the fault-seed. Owing to resource constraints in certain system such as embedded system, the run-time data logging can be incomplete, resulting in partial observation of software runs. A feature of our analysis is to localize the possible root cause in presence of such partial observability of data variables.