Introduction
As a consultant and trainer, I am often asked by my clients and students how to deal with automated acceptance tests. One common question is whether automated acceptance tests should use the graphical user interface (GUI) provided by the application.
Consider the Context
Like all testing questions, this is a question that can only be answered well when we consider context – the missions and the givens of the project and the product. We do that by asking more questions.
- What questions are we trying to ask and answer with automated tests?
- To what extent has testing been done at lower levels of the application?
- To what extent are we trying to confirm known behavior vs. trying to discover or investigate as-yet-unrecognized problems?
- What tools do we have available? What tools will we have to obtain?
- What is our team’s perceived level of experience and skills with the tools?
- Has the application been developed for testability? Does it have an easily scriptable interface below the GUI? Does it provide log files?
- Can we get changes to the product and support from the developers when the application is insufficiently testable (e.g. when HTML elements are missing id tags)?
- How might we value speed, precision, and volume of tests? Is the GUI the place where we will obtain optimal automation value for those considerations?
- To what extent do we need to model real users and observe what they observe? To what extent is automation capable of making such observations? What things will a conscious, conscientious human – a skilled tester – be likely to notice that automation might miss?
- How might we create or use oracles that will allow the tool to help us to recognize a problem? (For more on oracles see Doug Hoffman’s three part article “Using Oracles in Testing and Test Automation” (1))
- Is a need to vary the data that is being used in a productive way (systematically, pseudo-randomly, or randomly)?
- How might we leverage GUI automation to drive the application quickly to a point where human testers can take over?
- How might automated GUI tests drive us toward confirmation bias – that is, the tendency to design and execute tests that confirm existing beliefs about the product?
- How might automated GUI tests lead us towards automation bias – the tendency to view results from automated processes as superior to human processes.
This might look like a long list, but you don’t have to answer these questions in great detail, nor do you have to get each one of them exactly right. A couple of moments of reflection on each one – a couple of minutes altogether – is likely rather than forgetting to consider it at all. If you are uncertain or stuck, a few more minutes of investigation or exploration could provide a huge return on investment. As you are in the middle of the project, continue to ask these questions periodically as a means of ensuring that the value of what you’re doing exceeds the cost.
Experience and Observation
In general, my experience and observation has been that as testing gets closer to the GUI, the cost of automating tests increases while the value derived from automated tests decreases.
Automated tests at the unit level tend to be:
- Simpler to automate
- Easier to comprehend and maintain
- More subject to falsifiable assertions that machines can evaluate
- Appropriately specific, where specificity matters
- Revealing for problems that are easier to troubleshoot and debug
- More immediately responsive to developers (since the developers tend to be the ones writing and running them)
- Confirmatory, in a place where confirmation is more useful and important
Automated tests at the GUI level tend to be:
- Complex and difficult to program
- Hard to understand and maintain
- Inadequate for recognizing problems that humans can easily identify
- Overly specific in places where ambiguity can be tolerated by people
- Revealing for problems that are more difficult to troubleshoot and debug
- Less immediately responsive to developers
- Confirmatory, at a level where investigation and discovery are more important
Now: I’ve said that this is a question that we can only answer well in context – but there is another question entirely: what do we mean by “acceptance tests”, and depending on what we mean, do we want to automate them at all? You might like to check “User Acceptance Testing – A Context-Driven Perspective” for some thoughts on that (2).
Michael Bolton
Notes:
- Hoffman, Doug: “Using Oracles in Test Automation“. LogiGear Newsletter, 2006. Available free at:
Using Oracles in Testing and Test Automation (Part 1 of 3);
Using Oracles in Testing and Test Automation (Part 2 of 3);
Using Oracles in Testing and Test Automation (Part 3 of 3). - Bolton, Michael: “User Acceptance Testing – A Context-Driven Perspective”. Proceedings of the Pacific Northwest Software Quality Conference 2007, page 535.
