This is a popular question among practitioners and researchers. If we settle for a less-rigorous criterion, such as line coverage instead of branch coverage, we might miss something. Plus this question brings the focus back to the metric, which we do not want.
Which criterion to use depends on the context: what you are testing at that moment and how rigorous you want the testing to be. Structural testing is meant to complement specification-based testing. When you dive into the source code and look for uncovered parts, you may decide to use branch coverage for a specific if expression but MC/DC for another if expression. This makes the approach less systematic (and, therefore, more prone to errors and different developers using different criteria), but it is the most pragmatic approach I know. You may want to perform some risk assessment to determine how important it is to be thorough.
My rule of thumb is branch coverage: I always try to at least reach all the branches of the program. Whenever I see a more complicated expression, I evaluate the need for condition + branch coverage. If I see an even more complex expression, I consider MC/DC.
Leave a Reply