I find Spec # a very useful tool for constructing and analyzing state models. It is similar to C# in terms of flexibility, but has the domain specific aids on expects of a state modeling tool.

I have not used it for driving test automation. But I did use it one time to figure how many possible ways a bug could occur for a system we were running in production. I mapped out what we understood regarding the fundamental variables and transition rules, and then I created a rule that would conditionally format the boxes in the diagram as red if specific state conditions existed.

I used color because the STD generated by Spec # was huge – printed out it was four feet high and eight feet long and the text was still only about 2 points tall. But with color, one could still stand far away from that printout – or look at it on a screen – and easily draw conclusions like “Everything to the left of this branch is good, everything to the right hits the bug state, and this subset of branches never self-correct”

This really helped because we were then able to target the transition points from good to bad – and discover that out of this massive state diagram, there were only about 5 or so sequences that had unique bugs needing to be fixed to make the problem go away.

It may or may not be necessary to add all the actions that create “State X”. Traversals through an STD that land on the exact same state are theoretically equivalence classes to all transitions that follow. So long as you are not concerned with testing this equivalence (I like to partition my state testing – have one test that checks things like equivalence, another that deals with different things that happen after the state) you can just add the shortest route to getting that state – maybe even just assume in the model that the state exists a priori.

But the other thing to concern yourself with is saliency. For example, assume there is registry keys stored in the system somewhere and it can be in one of three values – so if it is in your STD as result of some action, there are (size of model * 3) states. Assume that there is no reason why “Action M” or any of its following transitions would consume or change that variable. If that variable is the only thing distinguishing the state in question, then there is diminished risk in skipping it in the STD.