Let's build an xUnit-style test runner for Go!

Michael Whatcott
Michael Whatcott
 • 
July 2, 2018
Why? What do you mean 'why'? Because we can! What's wrong with you?
Tags

Writing test functions in Go is easy:

package stuff

import "testing"

func TestStuff(t *testing.T) {
    t.Log("Hello, World!")
}

Running test functions is also easy:

$ go test -v
=== RUN   TestStuff
--- PASS: TestStuff (0.00s)
	stuff_test.go:6: Hello, World!
PASS
ok  	github.com/smartystreets/stuff	0.006s

Preparing shared state for multiple test functions is problematic. The usual recommendation is to use table-drive tests. But this approach has its limits. For us, xUnit is the ideal solution. It's simple, lightweight, and flexible. Wouldn't it be nice if we could define test methods on struct types and leverage common xUnit conventions like setups/teardowns, skipped tests, etc..? I'm thinking along these imaginary lines:

package stuff

import "testing"

// Define fields to manage system-under-test here (the fixture state).
type TestCase struct {
	*testing.T // Embedding *testing.T seems like a good idea for defining a test suite.
	sut *SystemUnderTest
}

// Perform setup actions here (instantiate test fixture state).
func (t *TestCase) Setup() {
	t.sut = NewSystemUnderTest()
}

func (t *TestCase) Test42() {
	if result := t.sut.Computation(42); result != 42 {
		t.Errorf("Got: [%d] Want: [%d]", result, 42)
	}
}

func (t *TestCase) Test43() {
	if result := t.sut.Computation(43); result != 43 {
		t.Errorf("Got: [%d] Want: [%d]", result, 43)
	}
}

The only problem is that the go test tool expects top-level functions, not methods on a struct type. And that's not going to change.

$ go test -v
testing: warning: no tests to run
PASS
ok  	github.com/smartystreets/stuff	0.006s

So, we need a way to connect a test function to methods on a struct type. And ideally, we could instantiate new instances of that type (with freshly initialized state) for each test method. Maybe a variation that leverages subtests would be closer to reality?

package stuff

import "testing"

func TestStuff(t *testing.T) {
	t.Run("Test42", new(TestCase).Test42)
	t.Run("Test43", new(TestCase).Test43)
}

// Define fields to manage system-under-test here (the fixture state).
type TestCase struct {
	sut *SystemUnderTest
}

// Perform setup actions here (instantiate test fixture state).
func (test *TestCase) Setup() {
	test.sut = NewSystemUnderTest()
}

func (test *TestCase) Test42(t *testing.T) {
    test.Setup()
	if result := test.sut.Computation(42); result != 42 {
		t.Errorf("Got: [%d] Want: [%d]", result, 42)
	}
}

func (test *TestCase) Test43(t *testing.T) {
    test.Setup()
	if result := test.sut.Computation(43); result != 43 {
		t.Errorf("Got: [%d] Want: [%d]", result, 43)
	}
}

That was certainly more effective:

$ go test -v
=== RUN   TestStuff
=== RUN   TestStuff/Test42
=== RUN   TestStuff/Test43
--- PASS: TestStuff (0.00s)
    --- PASS: TestStuff/Test42 (0.00s)
    --- PASS: TestStuff/Test43 (0.00s)
PASS
ok  	github.com/smartystreets/stuff	0.006s

But there are problems with this approach. Every time we define a new test method on the TestCase type we have to remember to register a subtest in the top-level test function. Oh, and did you notice how each test was calling the Setup method directly? This is something that should happen automatically if we're going to call this an xUnit-style test runner. It would be great if we could just call a method that points to our TestCase and iterates all test methods, calling Setup followed by a call to the test method.

From the calling side it could look something like this:

func TestStuff(t *testing.T) {
    xunit.RunTests(new(TestCase), t)
}

Notice we have to provide the *testing.T and an instance of our TestCase. The behavior defined in the mysterious xunit package would then find all the tests and run them. Impossible, you say? Not so! In fact, a draft implementation is trivial!

package xunit

import (
	"reflect"
	"strings"
	"testing"
)

func RunTests(fixture interface{}, t *testing.T) {
	fixtureType := reflect.TypeOf(fixture)

	for x := 0; x < fixtureType.NumMethod(); x++ {
		testMethodName := fixtureType.Method(x).Name
		if strings.HasPrefix(testMethodName, "Test") {
			// IMPORTANT: each test gets a new instance!
			instance := reflect.New(fixtureType.Elem())

			setupMethod := instance.MethodByName("Setup")
			callableSetup := setupMethod.Interface().(func())
			callableSetup()

			testMethod := instance.MethodByName(testMethodName)
			callableTest := testMethod.Interface().(func(t *testing.T))
			t.Run(testMethodName, callableTest)
		}
	}
}

This implementation makes a LOT of assumptions, lacks several features (like 'teardowns' and skipped tests) and isn't very robust, but hopefully you can see the emergence of an xUnit-style test runner. Most importantly, the tests are passing again:

$ go test -v
=== RUN   TestStuff
=== RUN   TestStuff/Test42
=== RUN   TestStuff/Test43
--- PASS: TestStuff (0.00s)
    --- PASS: TestStuff/Test42 (0.00s)
    --- PASS: TestStuff/Test43 (0.00s)
PASS
ok  	github.com/smartystreets/stuff	0.006s

Congratulations, you now possess a basic understanding of the inner workings of gunit! Stay tuned for a future post featuring a more in-depth look into xUnit-style testing in Go with gunit. In the meantime, feel free to kick the tires and fix things up a bit.


Source Code Download

Subscribe to our blog!
Learn more about RSS feeds here.
rss feed iconSubscribe Now
Read our recent posts
6 Ways Telecom & Internet Companies Leverage Address Data
Arrow Icon
In a recent webinar, we showed how address data tools help managers improve business performance in the following areas:Network PlanningAsset ManagementCustomer Acquisition and RetentionData BlendingMail and Parcel DeliveryFraud Reduction1. Network PlanningAccurate address data plays a critical role in overcoming network planning challenges. We presented solutions at each of the following steps in the network planning process:&nbsp;Clean existing address data with&nbsp;US Address VerificationKeep incoming data clean with&nbsp;US Address AutocompleteMap all addresses using our&nbsp;US Master Address List and hyper-accurate&nbsp;US Rooftop GeocodingAdd context to maps with&nbsp;enriched US Property DataAnalyze, get spatial insights, and make confident business decisions&nbsp;2.
A Fresh Start: Clean Out Your Computer Day
Arrow Icon
In a world where the lines between digital and physical spaces blur, the health of our computers has a massive impact on our productivity and overall ability to perform well at work. &nbsp;Clean Out Your Computer Day, observed on the second Monday of February, is the perfect opportunity to ensure that we declutter and optimize our virtual spaces. Just as we regularly clean our homes and offices, this day underscores the importance of maintaining a streamlined, efficient computer setup. The Digital Declutter: Where to StartStarting with the basics, sift through your files and programs.
Smarty's 2023 Address Data Deep Dive: Trends, Challenges, and Breakthroughs
Arrow Icon
We recently hosted a webinar where Brent Francom and Jeffrey Duncan, two of our Product Managers, talked about the lessons we learned and the improvements we've made in 2023. Here are the highlights. When you process so many addresses, you see things. We processed 327 billion+ addresses in 2023, equating to nearly 900 million per day or over 10,000 per second. When you process that many addresses, you find many funny and interesting ones. We keep a "Wall of Shame" with funny and malformed addresses that somehow validate.
Ready to get started?