Abstrakt
Dynamic analysis is a popular approach to detecting possible runtime errors in software and for monitoring program behavior, which is based on precise inspection of a single execution trace. It has already proved to be useful especially in the case of multithreaded programs and concurrency errors, such as race conditions.
Nevertheless, usage of dynamic analysis requires good tool support, e.g. for program code instrumentation and recording important events. While there exist several dynamic analysis frameworks for Java and C/C++ programs, including RoadRunner, DiSL and Valgrind, we were not aware of any framework targeting the C# language and the .NET platform.
Therefore, we present SharpDetect, a new framework for dynamic analysis of .NET programs - that is, however, focused mainly on programs compiled from the source code written in C#. We describe the overall architecture of SharpDetect, the main analysis procedure, selected interesting technical details, its basic usage via command-line, configuration options, and the interface for custom analysis plugins.
In addition, we discuss performance overhead of SharpDetect based on experiments with small benchmarks, and demonstrate its practical usefulness through a case study that involves application on NetMQ, a C# implementation of the ZeroMQ messaging middleware, where SharpDetect found one real concurrency error.