Experiments towards model-based testing using Plan 9: Labelled transition file systems, stacking file systems, on-the-fly coverage measuring

Axel Belinfante

    Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademic

    67 Downloads (Pure)


    We report on experiments that we did on Plan 9/Inferno to gain more experience with the file-system-as-tool-interface approach. We reimplemented functionality that we earlier worked on in Unix, trying to use Plan 9 file system interfaces. The application domain for those experiments was model-based testing. The idea we wanted to experiment with consists of building small, reusable pieces of functionality which are then composed to achieve the intended functionality. In particular we want to experiment with the idea of 'stacking' file servers (fs) on top of each other, where the upper fs acts as a 'filter' on the data and structure provided by the lower fs. For this experiment we designed a file system interface (ltsfs) that gives fine-grained access to a labelled transition system, and made two implementations of it. We developed a small fs that, when 'stacked' on top of the ltsfs, extends it with additional files, and an application that uses the resulting file system. The hope was that an interface like the one offered by ltsfs could be used as a general interface between (specification language specific) programs that give access to state spaces and (specification language independent) programs that use (walk) those state spaces like simulators, model checkers, or test derivation programs. Initial results (obtained on a less-than-modern machine) suggest that, although the approach by itself is definitely feasible in principle, in practice the fine-grained access offered by ltsfs may involve many file (9p) transactions which may seriously affect performance. In Unix we used a more conservative approach where the access was less fine-grained which likely explains why there we did not suffer from this problem. In addition we report on experiments to use acid to obtain coverage information that is updated on-the-fly while the program is running. This worked quite well. The main observation from those experiments is that the basic block notion of this approach, which has a more 'semantical' nature, differs from the more 'syntactical' nature of the basic block notion in Unix coverage measurement tools like tcov or gcov.
    Original languageUndefined
    Title of host publicationProceedings of the First International Workshop on Plan 9
    EditorsG. Guardiola, E. Soriano, F.J. Ballesteros
    Place of PublicationMadrid
    PublisherUniversidad Rey Juan Carlos
    Number of pages12
    ISBN (Print)84-690-2787-5
    Publication statusPublished - Dec 2006
    EventFirst International Workshop on Plan 9, IWP9 2006 - Madrid, Spain
    Duration: 4 Dec 20065 Dec 2006
    Conference number: 1

    Publication series

    PublisherUniversidad Rey Juan Carlos


    WorkshopFirst International Workshop on Plan 9, IWP9 2006
    Abbreviated titleIWP9


    • METIS-237676
    • EWI-8357
    • IR-66688

    Cite this