What's an SDET - II

Yesterday I discussed the same test triangle - and pointed out that the test focused solely on inputs and expected results, that it was a very dev-ish test. I'd like to explain why.

When you focus on inputs and expected results, I can almost see the code. It's something like this:

my $triangle_type = get_triangle_type($sidea, $sideb, $sidec);

A strong developer can test this function in several ways. He'll test the basic equivalence classes, the boundaries, maybe measure statement coverage - maybe even branch coverage, perhaps have a bunch of negative tests like entering words instead of numbers, or a "length" of a negative number. And, in the days of MS-DOS where you had a single user sitting at a single computer typing in one value at a time, that might be just fine.

In today's modern environments, that's only half the story. Because we'll take that simple program, wrap it in a web service, then create a web-based application that calls the web service. This "second half" has an entirely different set of potential problems:

- Does it render correctly in all browsers? IE6, IE7, FF2, FF3, Safari?
- Does it look pretty? Is the user interface usable?
- What happens if I resize the browser? Does the new rendering make sense?
- If I click tab through the various inputs instead of using the mouse, does the change in order make sense?
- If I press the "ENTER" or "RETURN" key, does that trigger the submit button?
- What happens if I click "submit" twice in a row - really fast?
- What happens if, after I click submit, I click the back button? Do I go back to the main screen or do I get one of those bizarre "This page was generated dynamically do you want to re-post?" error messages?
- What if I am visually impaired? Can I turn up the font or does the Cascading Style Sheet "lock" down the user experience? If I can crank up the font is it visually appealing?
- What if I am blind? Can I use the application by a tool for the blind like Lynx? Do all of the images have "alt=" tags?
- Is the web service reasonably fast? What if it's used by 100 users all at the same time? (Note: This was never a problem on MS-DOS, where you only had one user at a time)
- Can I run the application on my 1024x600 netbook? The ads said my netbook was good "for web surfing"
- Can I run the application on my Cell Phone?
- If I come in from a chinese, korean, or italian system but I know english, does the user experience make sense?
- What if I don't know English? Should our software be localized? If yes, what localizations?

You'll notice that none of those examples has anything to do with the core code; the web service could be completely correctly but the user experience still be buggy and unusable. This "second half" of the testing equation isn't about bits and bytes. It has much more to do with critical thinking than computer science - in fact, it is it's own separate and distinct discipline.

This is why I found the triangle example less than ideal; it focuses on one type of testing and completely ignores another. There is simply no way to ask it "what happens when I resize the browser?"

Hiring developers to be testers, you tend to get developer myopia - the focus on the "first half" - code, code, and statement coverage - and less focus on that second half. I don't think I need to names to say that we've all used applications that might have done exceedingly well on the first half of testing and yet failed miserably to provide a good user experience.

Now, the Microsoft Guys claim to be doing it right. They want Software Design Engineers in Test (SDETs) who can do *both* entry-level developing *and* critical investigation of software under test - and there are some people who fit into that category. But that's like saying you want excellent sprinters who are also world-class distance runners - while they do exist, there just ain't that many of those people on the planet. Those that are here can usually find gainful employment, and not all of them want to live in Washington State, China or India. The result is that, as an employer, you'll either have to (A) Pay an relative large sum for these people, (B) Have a bunch of open positions while you look for people with the right mix, or (C) Compromise, hiring people who are, for example, good devs you think might make good testers. This runs the serious risk of developer myopia.

Last time I checked (before the tech downturn), Microsoft has a few *hundred* open SDET positions. Given a choice of compromise and an HR department that won't allow you to pay a princely sum, that's probably the best choice.

I was discussing this with my colleague, James Bach, and he wrote something I would like to repeat:

The words that you quoted [Matt talking about MS's view of testers] represent an attitude that systematically misunderstands testing as a purely (or primarily) a technical activity the object which is to produce "test cases." I too had that attitude, early in my career. I grew out of it as I came to understand, through my experiences as a test manager, that a test team becomes stronger when it hosts a variety of people with a variety of backgrounds. Through the ordinary action of teamwork, a diverse group of thinkers/learners exploits the knowledge and skills of each for the benefit of all.

My attitude about testing is deeply informed by a study of cognitive psychology, which is the study of how people think, and epistemology, which is the study of how people can know what they know. ... When you approach testing not as hot-dogging with test tools or techniques, but rather as a process of human minds encountering human artifacts and evaluating them in human terms for other humans, you eventually realize that the testing process is harmed when any one point of view or way of thinking comes to dominate it.

I would like at least one programmer on my test team. Maybe a few. In some cases (such as in testing development tools) I will need everyone to be a programmer. However, I do not treat programming ability as the center of gravity for testing. Instead I look for rapid learning, high tolerance for uncertainty, and general systems reasoning. Even then, I don't need EVERYONE to be good at those things.

I'd use different rhetoric and be less critical, but I understand what James is saying. As testers, we tend to have developer-envy. The reality is that the two skills are separate, distinct, and complementary. (Unless, say, you are testing a compiler or a debugger. Are you?)

Now, can a non-developer-tester be a more effective by picking up a little code? Absolutely. I have an undergraduate degree in Math/CS and a Master's in CIS, of which I am extremely proud - not to mention a decade with a title of developer on my business card. AND it took me years to fight my way through developer myopia to see the whole picture of software testing.

In my experience, Developers tend to think in terms of automatable business processes - when exactly what needs to be done up front isn't clear, developers claim that the requirements are "inconsistent" and refuse to program.

The whole picture of testing might include some repeatable elements, but it also includes empirical processes - which adapt through learning as they happen. This is not a straightforward business process to automate. Developer-Envy doesn't help our craft, it hurts it.

That's just my opinion right now, I'm always open to changing it or presenting it more effectively.

... And with that, I'm off with my family to a week and half of vacation in the Caribbean. I welcome your flames, er, I mean comments.