Automated Debugging

Statistical Fault Localization

The idea here is that we assign scores to program statements based on their occurrences in passing and failing tests. Here, correlation relates to causation. If a statement is always executed during failed tests but not in passing tests, the statement is more likely to be related to a bug.

To produce these rankings:

• Given a test suite T
• $Score(s)=\frac{\frac{fail(s)}{allfail}}{\frac{fail(s)}{allfail}+\frac{pass(s)}{allpass}}$
• $fail(s)$ = # of failing executions in which s occurs
• $pass(s)$ = # of passing executions in which s occurs
• $allfail$ = Total # of failing executions
• $allpass$ = Total # of passing executions

See

• Delta Debugging