Eliminating exception handling errors with dependability cases: acomparative, empirical study

摘要

Programs fail mainly for two reasons: logic errors in the code andnexception failures. Exception failures can account for up to two-thirdsnof system crashes, hence, are worthy of serious attention. Traditionalnapproaches to reducing exception failures, such as code reviews,nwalkthroughs, and formal testing, while very useful, are limited inntheir ability to address a core problem: the programmer's inadequatencoverage of exceptional conditions. The problem of coverage might benrooted in cognitive factors that impede the mental generation (ornrecollection) of exception cases that would pertain in a particularnsituation, resulting in insufficient software robustness. This paperndescribes controlled experiments for testing the hypothesis thatnrobustness for exception failures can be improved through the use ofnvarious coverage-enhancing techniques: N-version programming, groupncollaboration, and dependability cases. N-version programming andncollaboration are well known. Dependability cases, derived from safetyncases, comprise a new methodology based on structured taxonomies andnmemory aids for helping software designers think about and improvenexception handling coverage. All three methods showed improvements overncontrol conditions in increasing robustness to exception failures butndependability cases proved most efficacious in terms of balancing costnand effectiveness