Automatically Inferring Structure Correlated Variable Set for Concurrent Atomicity Safety

Long Pang,Xiaohong Su,Peijun Ma,Lingling Zhao

DOI: https://doi.org/10.5121/ijsea.2013.4605

2013-01-01

Abstract:The atomicity of correlated variables isquite tedious and error prone for programmers to explicitly infer and specify in the multi-threaded program. Researchers have studied the automatic discovery of atomic set programmers intended, such as by frequent itemsetmining and rules-based filter. However, due to thelack of inspection of inner structure, some implicit sematics independent variables intended by user are mistakenly classified to be correlated. In this paper, we present anovel simplification method forprogram dependency graph and the corresponding graphmining approach to detect the set of variables correlated by logical structures in the source code. This approach formulates the automatic inference of the correlated variables as mining frequent subgra ph of the simplified data and control flow dependency graph. The presented simplified graph representation of program dependency is not only robust for coding style’s varieties, but also essential to recognize the logical correlation. We implemented our me thod and compared it with previous methods on the open source programs’ repositories. The experiment results show that our method has less false positive rate than previous methods in the development initial stage. It is concluded that our presented methodcan provide programmers in the development with the sufficient precise correlated variable set for checking atomicity .