Metamorphic relations to improve the test accuracy of Multi Precision Arithmetic software applications
Chittineni Aruna, R. Siva Rama Prasad · 2014
Recent advances in computing technologies are increasing the expectations of high accuracy and reliability from sophisticated arithmetic programs. Multi Precision Arithmetic (MPA) plays a vital role in majority of scientific applications, where the accuracy levels are more considerable and even a small mistake may misguide the downstream experimental results. Normal testing strategies rely on test oracles to predict the expected output to compare with target output. Testing of MPA programs is a crucial work with normal testing strategies, due to the complexity of generating oracles to verify the correctness of test outputs. In this paper we propose a novel software testing technique called Metamorphic Testing (MT), to test the non-testable MPA programs with the help of Metamorphic Relations (MRs) to alleviate the oracle problem.MT uses the data diversity technique to generate the follow-up test cases based on the existed successful test cases, which are low cost, scalable, efficient and leads to ‘N-Version Programming’. Experimental results are showing that our approach is identifying the hidden errors and improving the testing accuracy by finding more mutants.