A mixed approach to spectrum-based fault localization using information theoretic foundations
Year: 2013
Type: article
Abstract: Fault localization, i.e., locating faults in code, such as\nfaulty statements or expressions, which are responsible for observed\nfailures, is traditionally a manual, laborious, and tedious task.\nRecent years have seen much progress in automated techniques for fault\nlocalization. A particularly promising approach is to utilize program\nexecution spectra to analyze passing and failing runs and\ncompute how likely each statement is to be faulty. Techniques based\non this approach have so far largely focused on either using\nstatistical analysis or similarity-based measures, which have a\nnatural application in evaluating such runs. However, in spite of some\ninitial success, the current techniques lack the effectiveness of\nlocalizing the faults with a high degree of confidence in real\napplications.\n\nOur thesis is that information theoretic feature selection can\nprovide a basis for novel techniques that mix coverage of\ndifferent program elements for improving the effectiveness of fault\nlocalization using program spectra. Our basic insight is that each\nadditional failing or passing run can increase the information\ndiversity with respect to the program elements, which can help\nlocalize faults in code. For example, the statements with maximum\nfeature diversity information can point to the most suspicious lines\nof code. This dissertation presents a new fault localization approach\nthat embodies our insight and introduces Bernoulli divergence\nfor feature selection and uses it as the foundation for two novel\ntechniques: (1) mixing of branch and statement coverage information;\nand (2) varying of feature granularity from function-level to\nstatement-level. An experimental evaluation using a suite of subject\nprograms commonly used in evaluation of fault localization techniques\nshows that our approach provides an effective basis for fault\nlocalization.
Author: Shounak Roychowdhury
Language: English
Subfield: Software
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Award: G7160327619
OpenAlex ID: W1822654547
