Thoughts on Testing Strategies - Gnorion/BizVR GitHub Wiki

[6:34 PM] Mike Parish Since the dependencies are defined by the arrows (at least in the BCR view ) that's what you test after testing each table separately. Testing by rule seems a bit of overkill. But the user is free to do testing anyway he wants.

[6:36 PM] Mike Parish Final acceptance testing can then be done at the decision level - you should not need every possible test case if you have already done the prior steps - but there's no reason you couldn't set up a regression test with every combination - its would be valuable when you modify the decision.

[6:37 PM] Mike Parish The catch to testing every combination is that someone has to define what the expected results should be.

[6:39 PM] Mike Parish I suppose if you had a list of the rules as text (as taken from the original specification NOT from the model) an AI system might be able to figure out what the expected results should be. But I'm not sure I trust AI to do that, at least not yet.

[6:47 PM] Mike Parish In theory if you have "proved" every possible combination of inputs to table T1 always produces the correct output then when you combine T1 with T2 you only need to worry about T2's contribution (T1 is now a proven black box) (whether it feeds T2 or is fed by T2). Even if T2 sends the wrong data to T1 you know T1 will give you the correct response to that input. So you can limit your testing to ensuring that T2 is well behaved and gives correct outputs.

By taking this approach you only need to ADD the paths through each table rather than MULTIPLYING the paths through all the tables - that can be a huge difference. Makes an almost impossible task manageable.