Citation Context

... of v. Running several tests at once allows one to establish relationships between the executed code and the test outcome. For instance, one could isolate code that was only executed in failing tests =-=[15]-=-. This differential approach would also have isolated the erroneous code in our GCC example. Dynamic invariants [16] can be used to detect anomalous program behavior [17]. During execution, a tool che...

Citation Context

...y to lead to the defect than code that is executed in all runs. Harrold et al. found that failing runs tend to have unusual coverage [5]; such information can also be visualized to assist programmers =-=[8]-=-. However, applied to the Siemens suite, summarizing the coverage of failing runs either by intersection or union yields no useful results [12]. Slicing. Program slicing [15] yields the set of stateme...

Citation Context

...ave investigated ways of helping to automate this process of searching for faults. Some existing techniques use coverage information provided by test suites to compute likely faulty statements (e.g., =-=[1, 4, 5, 10, 12]-=-). Other techniques perform a binary search of the memory state using one failing test case and one passing test case to find likely faulty statements (e.g., [2, 15]). Still other techniques are based...

Citation Context

...t al. in [21], also work well for bug localization. Earlier work in [10] explores the relation between spectra differences and faults in regression testing. Similar information is later visualized in =-=[12]-=-, making it more accessible to the programmers. Recent researches by Renieris et al. show that spectrum differences are more effective for bug localization when it is applied along with nearest neighb...

Citation Context

... al describe a tool for defect localization called Tarantula, which uses color to visually map the participation of each statement in a program in the outcome of executing the program on a test suite =-=[16]-=-. Hildebrandt and Zeller describe a delta debugging algorithm that generalizes and simplifies failure-inducing input to produce a minimal test case that causes a failure [10]. Their algorithm, which c...

Citation Context

...any software analysis and visualization techniques have been developed for using information collected during execution to facilitate tasks such as program understanding and fault localization (e.g., =-=[9]-=-). SAV employs these techniques to help software producers to utilizes the possibly large amount of data that are collected from software instances. SAV may also be split in two submodules, SAVp and S...

Citation Context

...ler than either backward or forward dynamic slice. Harrold et al. [11] compared the spectra of passing and failing runs and found that failing runs tend to have unusual coverage spectra. Jones et al. =-=[16]-=- ranked each statement according to its ratio of failing tests to correct tests and used this information to assist fault location. Liblit et al. [19] describe a sampling framework and present an appr...

Citation Context

.... Since this technique relies on collecting state information while the program runs, the process of isolating the root cause for a failure could potentially havehigh runtime overheads. Jones et al. =-=[18, 19]-=- propose techniques that rank statements in a program based on their occurrence in failing and passing runs of the program. This ranking is subsequently used along with program visualization methods t...