At LShift, we tend to be big fans of functional programming, and in particular I’ve found ideas from languages like Clojure and Haskell do influence how I use more mainstream languages such as Ruby.
One technology that’s been useful to us on a current project is QuickCheck-alike for Ruby, Rantly. Briefly, rather than testing a module in your code by taking a set of (hopefully representative) examples of use and demonstrating that they produce the correct (usually pre-calculated) output, you can have the library generate input data and compare the results to a model.
So a fairly simple (and useful) example would be checking message serialisation. One problem we’ve had to solve is encode commands and responses to communicate with a remote service. Rather than just using Marshal or YAML to do this, we want to ensure that we can version messages correctly, and check that if say, the remote service changes it’s data model, then we can still interoperate with earlier versions of the client.
So, for our basic example, we’ll be encoding messages as JSON and checking two properties:
- That serialised messages are correctly representable as JSON without loss of information.
- That deserialising a serialised message gives us back what we put in.
The example code is up on github, as cstorey/rantly-example. Our first example is:
So, there’s a couple of things to note here
- First, we declare how to generate our input data. We’re using a lambda here because we need to have a different example object for each iteration that Rantly makes, not just per example (which rspec does by default)
- Rantly passes a context to the block as a parameter, and we use that to generate sub-elements of our input (in this case, we define the
uuidgenerator further up the file).
- Inside the test block, we pass our generator (declared as #1.) to
property_of, and then
checkrepeatedly calls the generator, and then passes that to our check block.
- And finally within the
checkblock, we can use ordinary rspec expectations, or even mocks if appropriate. Here, we are just testing for an identity (Hamsterdam, which we’re using to create value objects, gives us value-equality for free).
So, you might well be thinking that these tests are trivial, and in a sense they are, but there is definitely value in them. For example, in the case of schema versioning, then you can define generators for different versions of the
DoSomething message, and validate that say, we can parse them into an object that is meaningful within our domain model, and that they do still accurately reflect the original intent of the sender.
It’s also possible to build up more complex models as well, modelling operations as well as code (even up to the point of defining a little language for data persistence operations), but that’s something to look at next time.