Skip to content
Kevin Solorio edited this page Aug 22, 2017 · 8 revisions
Clone this wiki locally

In Discovery Testing, a Collaborator Object describes any unit of code whose responsibility is to orchestrate the invocation of other units, as opposed to containing non-trivial enumeration or branching logic.

When practicing Discovery TDD, a developer typically starts with a new collaboration test and asks "if several components could be composed to solve this problem, what would those components be?"

Example walkthrough of a collaboration test

Let's look at an example in JavaScript. Suppose we're building a treehouse. We might start with a test that invokes this function under test:

var subject = new TreeHouseBuilder()

var result = subject.build()

assert(result instanceof TreeHouseWithRailings)

It would be foolish to try to implement everything to do with building a treehouse inside of a single method. A collaboration test doesn't presume we know how to best do the job yet, but helps us focus on what building a treehouse would entail. Brainstorming (and reading up on treehouses), we break it up into a few categories:

var chooseTree
var buildPlatform
var addRailings

Mercifully, it's not our job at this point to answer how we'll choose a tree, design & build a platform, or install railings. All we need is a reasonable confidence that these three dependencies, if implemented well, would accomplish the task.

Of course, that won't work, because the three dependencies don't exist yet, so neither our test nor its subject have a reference to them. One option would be to pause our work on this test here, go implement those dependencies, and then finally implement the test. However, in Discovery Testing we'd prefer to prove out the subject's interaction with these dependencies before investing time in them. So, what do we do? This is where test doubles or "mocks" come into play.

Here's how that test might look if we used test doubles in place of real functions, then passed them to our subject (using testdouble.js to illustrate:

var chooseTree = td.function('chooseTree')
var buildPlatform = td.function('buildPlatform')
var addRailings = td.function('addRailings')
var subject = new TreeHouseBuilder(chooseTree, buildPlatform, addRailings)

var result = subject.build()

assert(result instanceof TreeHouseWithRailings)

[Of course, passing function references around like this seems a bit ham-fisted, which is why testdouble.js provides less obtrusive ways to replace dependencies with test doubles.]

If we ran this test, even in its incomplete state, we'd get several rounds of initial feedback via messages like TreeHouseBuilder is not defined, then (after defining it) subject.build is not a function, then (after defining it) TreeHouseWithRailings is not defined. Once clearing these messages, we get to what might be called the "logical" or "intended" failure of AssertionError: false == true.

Now we have to think about how the subject would invoke these three things to ultimately return a TreeHouseWithRailings. We accomplish this with configuring stubbings on the test doubles.

A stubbing looks like this:

var tree = new Tree()
td.when(chooseTree()).thenReturn(tree)

Where td.when is used to configure the test double chooseTree to return a particular tree when it's invoked with no arguments.

Filling out the rest of the stubbing setup also typically shakes out several value types, namely Tree, TreeHouse:

var chooseTree = td.function('chooseTree')
var buildPlatform = td.function('buildPlatform')
var addRailings = td.function('addRailings')
var subject = new TreeHouseBuilder(chooseTree, buildPlatform, addRailings)
var tree = new Tree()
td.when(chooseTree()).thenReturn(tree)
var treeHouse = new TreeHouse()
td.when(buildPlatform(tree)).thenReturn(treeHouse)
var treeHouseWithRailings = new TreeHouseWithRailings()
td.when(addRailings(treeHouse)).thenReturn(treeHouseWithRailings)

var result = subject.build()

assert(result instanceof TreeHouseWithRailings)

Additionally, we can make the assertion more specific now that we have greater control over how the test doubles behave. Instead of a heuristic assertion based on instanceof:

assert(result instanceof TreeHouseWithRailings)

We can assert that the tree house with railings that is returned is the exact one that addRailings has been stubbed to respond with when passed the exact treehouse that was built:

assert.equal(result, treeHouseWithRailings)

Other examples

Here is another example of a collaboration test: