UI testing – interview with Alex Imrie

July 6th, 2009

After talking with Alex Imrie about usability, it’s time to ask her a few questions about one of the tools her company is working on – GUIDancer. Following a similar interview with Alex Ruiz, creator of FEST, this interview delves deeper into the subject of testing desktop and web UIs.
alexandra_imrie_thumbTell us a little bit about yourself

I’m Alex Imrie, and I work at BREDEX GmbH in Braunschweig, Germany. I have various roles at the company including Marketing, customer demonstrations, training and support as well as documentation and conceptual design for some of our software. I also do some testing with our automated test tool, GUIdancer.

What is GUIdancer?

GUIdancer is a testing tool for automating functional tests through the GUI. In essence, tests that are usually performed manually can be automated with GUIdancer. We currently support applications with Java (Swing, SWT/RCP) and HTML user interfaces. GUIdancer is a black-box tool and differs from other similar tools in that it uses the keyword-driven approach to testing. Keyword-driven testing is a method which is very close to the principles of software development without actually requiring that code be written. Because GUI tests consist of the same repeated actions, there is a focus on reusability. Tests can be created from a running application or parallel to software development, independently of an application under test, from a library of actions by drag & drop. Each module is named according to the actions it executes, and can be reused (referenced) throughout the test. This reusability means that tests grow quickly and are easy to maintain because central changes update all the instances where a module was reused.

Why did you decide to focus mainly on testing the Java UIs?

Since Bredex was founded in 1987, most of our projects have involved user interfaces, so there has been a focus on GUI testing since the very early days. From 1995, we specialised in Java, so when we decided to write our own test tool, the choice was obvious which technology we were going to start with. The irony was, the first toolkit we supported was Swing, and GUIdancer itself is written in RCP. Since the release of version 2.0 we’ve been able to test GUIdancer with GUIdancer, and we also added the support for HTML testing. The architecture of GUIdancer means that any interface can be tested; it’s just a question of seeing which direction we plan to go in next.

UI testing doesn’t seem to receive its share of the “limelight”, even in the currently popular test driven development paradigm. Is it too difficult or is it just seen as less important?

I think that there is certainly the perception that it is too difficult. A lot of people have been burned by failed attempts or have started with the wrong expectations of functional test automation. UI testing, even in the test driven development paradigm, is by no means impossible. It’s just important to keep realistic aims in mind – automation of repetitive tasks first, for example, or simply having stable regression tests for core features that run as soon as a new piece of the software becomes available. Sure, there are difficulties with functional testing, but these can be avoided by taking the time to identify clear goals and design and plan the tests. With GUIdancer, we see that the tester to developer ratio on a project is as low as 1:10, so the support in the tool for structure and planning pays off well.

Having said that, I think that the importance of UI testing does also tends to be forgotten. It is certainly possible to test a great deal with JUnit, for example, and such tests are incredibly important. However, there is also the need to test the application from the user’s perspective. Can the simplest use case be easily completed via the GUI? Can the application be brought into an irregular state by user actions? On another important level – does the application even do what the customer ordered? These are areas where UI testing really shines, and where JUnit alone doesn’t suffice.

Should testing infrastructure be part of a UI toolkit or is this better left to interested third parties?

As an interested third party, I fear my answer may be somewhat biased! There is a certain charm to a centralised test framework, but I think that third parties are better positioned to know how the toolkit is used in practice and what deviations from the standard are common. There is also the argument that different teams and organisations use different approaches to testing and even different skillsets in the testing team. In all aspects, I believe, there is simply too much variation for a central infrastructure.

Have you looked into the scenegraph approach to building UIs in JavaFX? Does it present significant challenges to existing Java UI testing toolkits?

We haven’t looked into it in detail, but I think the scenegraph approach could pose certain challenges, yes. The animation aspects would mean that timing and movement have to be considered in the tests – there would have to be some pretty good synchronization to ensure robustness, I think.

Do you see desktop applications as a dying breed, with all the significant advantages of browser hosted solutions?

Web applications are certainly very popular and I doubt that this popularity is going to end soon. I also doubt, though, that they will completely replace desktop applications. There is still a strong demand for local applications – which do not need the various capabilities (and complications) that come with browser solutions. One reason I quite like desktop applications is because they are generally more ergonomic and usable. I think web applications have a lot of catching up do to in this respect. Local applications have better dialog support and don’t need to be manually refreshed quite as often. I also see many web applications that pack way too many things on one page so that scrolling (in all directions) is unavoidable.

Would you like to share the future plans for GUIdancer?

We are working towards the release of version 3.1 at the moment, which will be released in the second week of July. This release will see the introduction of automated testing for GEF applications, which I’m very excited about! GUIdancer 3.1 will also be compatible with the Eclipse Galileo release. There will be a few more goodies too, like more supported actions on tables and better support for native dialogs. 3.2 and 4.0 are already being planned, with more toolkits and browsers on the list, as well as a test execution manager and more possibilities to manage test data.