There are several commercial products that use proprietary languages, which typically look like a wrapper around (some proprietary extension of) the standard SQL language. Examples of these languages include ABAP, Informix, XBase++, SQR and so on. These application are difficult to analyze not only because it is hard to model the semantics of the underlying database systems but also because of the lack of standard tools for analysis. One naive way to analyse such programs is to collect dynamic trace using proprietary debuggers and run the analyses on the trace. However, this form of dynamic trace collection can be a severe performance bottleneck. In this paper, we present our experience with building a framework to help in efficient program analysis in the context of ticket resolution for ABAP programs.

In our framework, we first translate the given ABAP programs to semantically equivalent annotated Java programs. These Java programs are then executed to generate the required dynamic trace. Our framework allows the plugging of off-the-shelf static analysis tools (applied on the Java programs) and dynamic trace analysis tools (on the generated trace) and maps the results from these analysis tools back to the original ABAP programs. One novel aspect of our framework is that it admits incomplete ABAP grammar, which is an important aspect when dealing with proprietary languages where the grammar may not be publicly available. We have used our framework on several benchmarks to validate the translation, and establish the efficiency and the utility of our instrumented Java code along with the collected trace.