Sunday, 7 August 2011

Reflections on "Done": Regression & Interaction Testing

" #softwaretesting #testing "

If you haven't read it then go and read Michael Bolton's post and comments regarding the usage of "done", here.

From some of the comments, discussions and opinions elsewhere there is a notion that "done" is the complete package. This was the point of Michael's post. But with each sprint, delivery or increment the product evolves - and so any definition of done will be susceptible to problems with definition.

It's very easy to have a mission that includes "existing features work as before" - but there is also a risk of missing something here….

The product evolves - interactions between features and within the system change. Not all of this can be anticipated in advance.

So, however you might define "done" for your delivery there will (inevitably) be an area of unknowns that may* be investigated.

It's not as simple as saying this is covered by the exploratory testing (ET) part of the sprint. The ET may be for the new feature, but there is also an element that could* be devoted to learning about interactions between the feature and existing product features or components, and even understanding how the whole behaves (again**).

Of course, a good first step here is to separate out the ET missions for the new feature and the ET missions for interactions and existing features.***

Some of this might be covered in some "regression" and characteristics measurement and monitoring. But the problem with "regression" is that it doesn't necessarily have an element of learning about how some existing test suite (especially applicable to automated suites) works with the "new" (evolved) system.

An automated regression suite usually has a notion of "this is a fixed reference point" - this script should**** usually work. But the testing that is missing is usually the evaluation of "if and how" it works in the new system. This evaluation is commonly limited to looking at failures and how they should be fixed.

Some shops wrap this up as a "regression" element into the increment's mission (or even definition of done) - but wrapping things up (implying) in some other activity is exactly the original problem with "Done" - it doesn't necessarily reflect the focus of the problem in front of you at a given time.

Now, being able to reflect "exactly" the problem in front of us is something that can't be done "exactly" - that's why we need testing to help evaluate. So, dimensioning (or estimating) an activity should be wary of this.

Ultimately, a lot of this comes down to good testing - separating the assumptions (and implications) from the problem in front of you and illuminating these assumptions and implications with the stakeholder as early as possible.

Good testing starts early!

* even "should" or "must" depending on the context
** understanding how the system behaves is an iterative task.
*** Note, this is all depending on the specific context - it's not automatic to say "do this for every iteration" - each delivery must be assessed on its needs and usage.
**** "should" is one of the most over-and-misused words within software development!

No comments:

Post a Comment