I’ve been delving into the more technical side of testing, that of using tools to exercise code precisely rather than galumphing through UI, which really is my preferred method. Our mobile app uses a lightweight service to connect to the other moving pieces at the company and our vendors, so we can test those calls directly through SOAP, which, for you non-software people out there, is also called API testing.
The developers of our software took kind of a blanket approach to the envelopes for the service calls, so there are tons of superfluous fields for each of the calls, and it’s kind of a guessing game as to which are required for any particular service call. My developers didn’t create all the service calls, so I’ve actually been able to figure things out and share my envelopes with them (so, you know, I feel pretty cool).
Once I figure out which fields are necessary to get a call working, then it’s time to figure out an assertion. These are what the testing software checks against to make sure things are working appropriately. In our testing, they’re usually related to the content of a field, whether it is a “success” or “failure” message, or, in a case I was working on recently, whether the name of a document started with a specific year or not.
With this case, we modified the call to pull documents one year at a time, so the call for each year needed to return only documents for that year and none others. I ended up doing a check that the count of docs that started with something other than the year I was looking for was 0. This may sound simple, but it took quite awhile to get the xpath right, and I had some (a lot of?) help from a developer.
The SOAP tests are nothing without the correct assertions. You can go ahead and get all the information you want, but without an automated check on the information, the tests have very little advantage over UI testing in terms of speed, though they do have the ability to cut down on some of the noise that is inevitable with UI testing.
API testing is another tool in the box. It cannot be the only thing, and it should not be ignored. It’s great for getting precise information, for making sure you exercise calls at a lower level, and it can be faster than UI testing, but it is not a substitute for a full complement of testing, including exploratory testing (galumphing), scripted tests run manually, and scripted tests run through automation. And, of course, if your developers unit test their code, that helps too. 🙂