Abstract: Software fault localization and understanding are key steps in software debugging,while the way that debuggers use suspicious scores calculated by coverage-based fault localization from high to low to analysis every program statement does not correspond to real debugging.To select test executions which are more helpful for fault understan-ding,especially the failed executions which lead to program failure,and help debuggers conduct differentiation analysis dynamically,three prioritization strategies which are based on coverage rate of high suspicious scores,fault exposing potential and discrete characteristic of suspicious scores of covered statements were presented respectively for failed executions,and weighted cosine similarity matching was presented for passed executions.These failed test execution prioritization strategies and random sorting were compared in general fault localization techniques.And results show that the selection based on discrete characteristic of suspicious scores of statements covered by test executions has stronger positive influence and weaker negative influence on the change of fault understanding expense before and after selection,can understand more faults under same expense,and is more helpful for the application of fault localization results to fault understanding.

Key words: Software fault localization,Fault understanding,Coverage analysis,Test execution prioritization,Discrete characteristic