1000 conditions per decision ought to be enough for anybody
Modified condition/decision coverage (MC/DC) testing requires testing, among other things, that each condition in a decision takes every possible outcome, and independently affects the outcome of the decision.
Coverage analysis tools test for MC/DC by observing results when the value of each condition is changed independently from the other conditions in the decision (though, with some clever logic, masking MC/DC can reduce overheads in some cases).
Most coverage analysis tools support what they deem to be a "reasonable" number of conditions per decision (our customers report that limits of
20 conditions per decision are common). But what if one of your decisions exceeds this arbitrary limit? You could:
- Change your code (but why should you?)
- Perform MC/DC analysis by hand (how many conditions? 50? good luck!)
Why do you support so many conditions per decision?
Because we can, and because your code may include giant decisions (we aren't judging).
Surely, nobody would write a decision that big, right?
While you're unlikely to write something like this by hand, if you're using model driven development (MDD), your automatically generated code may include decisions with a large number of conditions (for example code for aggregating large numbers of boolean signals).
Why only 1000 conditions?
We can support more than 1000 if you really need it! (but we'd be curious to see the code that requires this).
What does a RapiCover report with such large decisions look like?
I'm glad you asked. See Figure 1 below. Note that you can scroll to view the full file in the source code view.
Figure 1. RapiCover report from a decision with almost 1000 conditions
The example in Figure 1 shows a function that aggregates hundreds of boolean signals (just short of
1000) into a "warning light" signal. This is a bit like the piece of software that lights up the engine light on your dashboard. There are many error signals that can cause this light to come on and they can be all listed in a large
or boolean expression making a very large MC/DC decision.
The example report above shows the RapiCover MC/DC view with the coverage status of close to
1000 conditions in both the table view and the colorized source code view.
In the source code view, you can see a few coverage holes. For example, while various subsystems are missing a few tests to cover all their signals, the Turbolift subsystem has not been tested at all, and all of the GravityGenerator signals have been addressed using justifications rather than tested.
As a final note, it is interesting to see that Ada is still a popular choice for Aerospace software development in the 23rd century - as shown by this example.
Want to read something more in-depth? Our white papers examine common challenges and solutions in critical software verification: