One of my current responsibilities is to find ways to automate as much as practical of the ‘testing’ of the user experience (UX) of complex web-based applications. In my view, full test automation of UX is impractical and probably unwise, however we can use automation to find potential problems[1] in UX even of rich, complex applications. I, and others, are working to find ways to use automation to discover various types of these potential problems. Here’s an overview of some of the points I made. I intend to extend and expand on my work in future posts.

In my experience, heuristics are useful in helping identify potential issues. Various people have managed to create test automation that essentially automates various heuristics.

Examples of pushing the boundaries

You might notice that all the examples I’ve provided are available as free opensource software (FOSS). I’ve learnt to value opensource because it reduces the cost of experimentation and allows us to extend and modify the code e.g. to add new heuristics relatively easily (you still need to be able to write code, however the code is freely and immediately available).

Automation is (often) necessary, but not sufficient

Automation and automated tests can be beguiling, and paradoxically increase the chances of missing critical problems if we chose to rely mainly or even solely on the automated tests. Even with state of the art (the best we can do across the industry) automated tests I still believe we need to ask additional questions about the software being tested. Sadly, in my experience, most automated tests are poorly designed and implemented, which increases the likelihood of problems eluding the automated tests.

Here are 2 articles which describe some of the key concerns.

The first describes how people can be biased into over-reliance on automation. It is called “Beware of Automation Bias” by M.L. Cummings, in 2004. The article is available online at http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.91.2634&rep=rep1&type=pdf

The second helped me understand where testing helps us work out which questions to ask (of the software), and that we need to use a process to identify the relevant questions. The article is called 5 Orders of Ignorance, by Phillip G Armour, CACM 2000 http://www-plan.cs.colorado.edu/diwan/3308-s10/p17-armour.pdf

Note: the essence of this material was presented as a lightning keynote at the Belgium Testing Days conference on 15th February 2011

[1] potential problems is one term I use to avoid getting into arguments about whether a problem is a bug or not. I prefer to use the term ‘undesirable effects’ since software (and things in general) may meet the requirements but still have undesirable effects. Here I’m happy to focus on potential problems; perhaps I’ll write a post on the topic of undesirable effects soon…

Introduction

Lots of companies, and teams have a burden of unreliable, problematic, automated tests that are troublesome and time-consuming to maintain. We need ways to address these poor-quality tests, ways to recover from the current unhealthy situation to a healthy environment where the tests are reliable, trustworthy, and easy to maintain as the underlying software changes.

For new test automation projects and teams, with the ‘luxury’ of starting fresh, you might also find these topics salutary, advance warning of a situation you might end up in if you don’t apply good test automation practices from the outset. Don’t say you weren’t warned 🙂

Here are the initial topics I’m going to cover

• Re-engineering and Refactoring automated tests
• Understanding Critical Success Factors for test automation
• Applying Design Patterns
• Design and Structure your test automation code
• Coping with large volumes of ‘legacy’ and ‘broken’ tests (including record & playback)
• Remove boiler-plating or dumbed-down interfaces
• Making the interaction with web pages resilient and robust (using IDs, working with developers, etc).
• Don’t be fooled (again) avoid being beguiled by automated tests
• Sunk by dependencies (e.g. on live back-end servers). Coping with Environmental Issues.
• Slow tests
• Modelling Techniques
• The intersection of automation and in-person testing
• Three possible outcomes of a test
• Writing readable code
• Stringing tools together
• Patterns for: Data creation, reuse, sharing and cleanup
• Designing tests to safely run in parallel

One of my aims is to create a useful, succinct guide to help you, and others, create and establish useful, readable and maintainable automated tests for your software projects. I’m drawing on the work, experience and expertise of various people in the software testing communities, and welcome your input and ideas.

“Sufficient to the purpose of Test Automation”

“Sufficient to the purpose”

Examples of Lean Test Automation

Here are a couple of personal examples where I managed to use test automation to find significant issues that the development teams then fixed (one measure of value of the test automation).

1. Writing a short script in Perl, about 10 lines of code, that helped expose a fundamental security flaw in a nationwide system, back in 2005. The work was presented at CAST 2006 in Indianapolis, USA. The proceedings used to be available from http://www.associationforsoftwaretesting.org/ but they’ve reorganised the site and don’t seem to have it available currently. I’ll see if I can make it available again.
2. Writing about 50 lines of Java to dynamically navigate web pages and web applications. These tests found several significant bugs on a variety of projects at work. The code is freely available at http://code.google.com/p/web-accessibility-testing/

In both cases, the code was relatively simple and relied on existing libraries to interact with the systems being tested. The effort to write the code was low (hours) compared to the value running the tests provided.

Further Reading