Are Mock Objects Too Powerful?
Lately at work we’ve been working towards test driven development and not just “having a lot of tests.” For pretty much anyone who has written unit tests for anything with any complexity, you know that’s a lot harder than it sounds. You need to be able to test certain components in an isolated fashion and the bits that you need to integrate with may not actually be written yet.
To get around this, you generally end up writing a lot of test and helper classes to stub in the functionality your component interacts with, but that’s a heck of a lot of work. In some cases, you might have to drop live configuration files into the system to get things to work correctly, you might need to craft some dynamic logic into your test classes… it’s a big pain.
Some folks choose to architect their components to be easier to test. This usually implies there are more publicly exposed methods than you might normally have so that certain internal properties can be checked on, substituted in, or otherwise dealt with in a test environment. It also means there are a lot more moving parts - interfaces for “plugging in” components that wouldn’t normally be there except for the need to swap in at test time. We’ll call that “designing for testability.”
Unfortunately, much of what I work on has the API as a deliverable. Which is to say, I can’t just have a load of exposed public methods floating out there solely to support my tests. I can’t “over-architect” the usage of the components because part of the goal is to make the components simple to use. Instead of designing for testability, we have to test what’s designed.
The problem, then, is how to “plug in” or stub out things in testing to isolate the component being tested? Enter mock object frameworks.
Mock object frameworks allow you to do that sort of thing on the fly. You can say “give me a mock data provider and whenever anyone queries it for data, have it return this data set here.” It’s a really nice, simple way of doing things that doesn’t require you to bloat your design just so you can test it.
Okay, so that’s your quick “mock objects” intro. The question I’m leading to is: mock objects are very powerful. You can do a lot of stuff with them. So much, that if you aren’t really paying attention, you could very well mock your way into invalid tests. The question on the table, then, is “are mock objects too powerful?”
This is actually an ongoing debate at work as we investigate different mock object frameworks. If we end up with a site license for a fairly powerful mock object framework, what’s to stop an untrained developer from misusing it and giving us a false sense of quality by writing invalid tests?
My view: it’s a tool, like a screwdriver or a hammer. Or matches. If you don’t know what you’re doing with the matches, you’re going to get burned. If you know what you’re doing, matches can be very helpful. It all comes down to education. People just need to be smart enough to know when they’re not smart enough to start using the tool and get educated before picking it up and heading down that path. I don’t think “people might misuse it” is a convincing enough argument to not use the tool. I might also consider that if it’s worrisome to a particular team or project, the folks overseeing that team or project need to pay attention and ensure the right tools are being used by the right people for the job.
Besides, there are so many other ways the uninitiated can mess up production code, somehow I think “using a mock object framework improperly in testing” doesn’t qualify on the top 10 threat list.