Golang basics - writing unit tests

In the previous post titled "Grab JSON from an API" we explored how to interact with a HTTP client and parse JSON. This post is a continuation of that theme, which covers unit testing.

1. Testing in Go

Go has a built-in testing command called go test and a package testing which combine to give a minimal but complete testing experience.

The standard tool-chain also includes benchmarking and statement-based code coverage similar to NCover (.NET) or Istanbul (Node.js).

Share & follow on Twitter:

1.2 Writing tests

Unit testing in Go is just as opinionated as any other aspect of the language like formatting or naming. The syntax deliberately avoids the use of assertions and leaves the responsibility for checking values and behaviour to the developer.

Here is an example of a method we want to test in the main package. We have defined an exported function called Sum which takes in two integers and adds them together.

package main

func Sum(x int, y int) int {  
    return x + y
}

func main() {  
    Sum(5, 5)
}

We then write our test in a separate file. The test file can be in a different package (and folder) or the same one (main). Here's a unit test to check addition:

package main

import "testing"

func TestSum(t *testing.T) {  
    total := Sum(5, 5)
    if total != 10 {
       t.Errorf("Sum was incorrect, got: %d, want: %d.", total, 10)
    }
}

Characteristics of a Golang test function:

  • The first and only parameter needs to be t *testing.T
  • It begins with the word Test followed by a word or phrase starting with a capital letter.
    • (usually the method under test i.e. TestValidateClient)
  • Calls t.Error or t.Fail to indicate a failure (I called t.Errorf to provide more details)
    • t.Log can be used to provide non-failing debug information
  • Must be saved in a file named something_test.go such as: addition_test.go

If you have code and tests in the same folder then you cannot execute your program with go run *.go. I tend to use go build to create a binary and then I run that.

You may be more used to using the Assert keyword to perform checking, but the authors of The Go Programming Language make some good arguments for Go's style over Assertions.

When using assertions:

  • tests can feel like they're written in a different language (RSpec/Mocha for instance)
  • errors can be cryptic "assert: 0 == 1"
  • pages of stack traces can be generated
  • tests stop executing after the first assert fails - masking patterns of failure

There are third-party libraries that replicate the feel of RSpec or Assert. See also stretchr/testify.

Test tables

The concept of "test tables" is a set (slice array) of test input and output values. Here is an example for the Sum function:

package main

import "testing"

func TestSum(t *testing.T) {  
    tables := []struct {
        x int
        y int
        n int
    }{
        {1, 1, 2},
        {1, 2, 3},
        {2, 2, 4},
        {5, 2, 7},
    }

    for _, table := range tables {
        total := Sum(table.x, table.y)
        if total != table.n {
            t.Errorf("Sum of (%d+%d) was incorrect, got: %d, want: %d.", table.x, table.y, total, table.n)
        }
    }
}

If you want to trigger the errors to break the test then alter the Sum function to return x * y.

$ go test -v
=== RUN   TestSum
--- FAIL: TestSum (0.00s)
    table_test.go:19: Sum of (1+1) was incorrect, got: 1, want: 2.
    table_test.go:19: Sum of (1+2) was incorrect, got: 2, want: 3.
    table_test.go:19: Sum of (5+2) was incorrect, got: 10, want: 7.
FAIL  
exit status 1  
FAIL    github.com/alexellis/t6 0.013s  

Launching tests:

There are two ways to launch tests for a package. These methods work for unit tests and integration tests alike.

1) Within the same directory as the test:

go test  

This picks up any files matching packagename_test.go

or

2) By fully-qualified package name

go test github.com/alexellis/golangbasics1  

You have now run a unit test in Go, for a more verbose output type in go test -v and you will see the PASS/FAIL result of each test including any extra logging produced by t.Log.

The difference between unit and integration tests is that unit tests usually isolate dependencies that communicate with network, disk etc. Unit tests normally test only one thing such as a function.

1.3 More on go test

Statement coverage

The go test tool has built-in code-coverage for statements. To try it with out example above type in:

$ go test -cover
PASS  
coverage: 50.0% of statements  
ok      github.com/alexellis/golangbasics1  0.009s  

High statement coverage is better than lower or no coverage, but metrics can be misleading. We want to make sure that we're not only executing statements, but that we're verifying behaviour and output values and raising errors for discrepancies. If you delete the "if" statement from our previous test it will retain 50% test coverage but lose its usefulness in verifying the behaviour of the "Sum" method.

Generating an HTML coverage report

If you use the following two commands you can visualise which parts of your program have been covered by the tests and which statements are lacking:

go test -cover -coverprofile=c.out  
go tool cover -html=c.out -o coverage.html  

Then open coverage.html in a web-browser.

Go doesn't ship your tests

In addition, it may feel un-natural to leave files named addition_test.go in the middle of your package. Rest assured that the Go compiler and linker will not ship your test files in any binaries it produces.

Here is an example of finding the production vs test code in the net/http package we used in the previous Golang basics tutorial.

$ go list -f={{.GoFiles}} net/http
[client.go cookie.go doc.go filetransport.go fs.go h2_bundle.go header.go http.go jar.go method.go request.go response.go server.go sniff.go status.go transfer.go transport.go]

$ go list -f={{.TestGoFiles}} net/http
[cookie_test.go export_test.go filetransport_test.go header_test.go http_test.go proxy_test.go range_test.go readrequest_test.go requestwrite_test.go response_test.go responsewrite_test.go transfer_test.go transport_internal_test.go]

For more on the basics read the Golang testing docs.

1.4 Isolating dependencies

The key factor that defines a unit test is isolation from runtime-dependencies or collaborators.

This is achieved in Golang through interfaces, but if you're coming from a C# or Java background, they look a little different in Go. Interfaces are implied rather than enforced which means that concrete classes don't need to know about the interface ahead of time.

That means we can have very small interfaces such as io.ReadCloser which has only two methods made up of the Reader and Closer interfaces:

        Read(p []byte) (n int, err error)

Reader interface

        Close() error

Closer interface

If you are designing a package to be consumed by a third-party then it makes sense to design interfaces so that others can write unit tests to isolate your package when needed.

An interface can be substituted in a function call. So if we wanted to test this method, we'd just have to supply a fake / test-double class that implemented the Reader interface.

package main

import (  
    "fmt"
    "io"
)

type FakeReader struct {  
}

func (FakeReader) Read(p []byte) (n int, err error) {  
    // return an integer and error or nil
}

func ReadAllTheBytes(reader io.Reader) []byte {  
    // read from the reader..
}

func main() {  
    fakeReader := FakeReader{}
    // You could create a method called SetFakeBytes which initialises canned data.
    fakeReader.SetFakeBytes([]byte("when called, return this data"))
    bytes := ReadAllTheBytes(fakeReader)
    fmt.Printf("%d bytes read.\n", len(bytes))
}

Before implementing your own abstractions (as above) it is a good idea to search the Golang docs to see if there is already something you can use. In the case above we could also use the standard library in the bytes package:

    func NewReader(b []byte) *Reader

The Golang testing/iotest package provides some Reader implementations which are slow or which cause errors to be thrown half way through reading. These are ideal for resilience testing.

1.5 Worked example

I'm going to refactor the code example from the previous article where we found out how many astronauts were in space.

We'll start with the test file:

package main

import "testing"

type testWebRequest struct {  
}

func (testWebRequest) FetchBytes(url string) []byte {  
    return []byte(`{"number": 2}`)
}

func TestGetAstronauts(t *testing.T) {  
    amount := GetAstronauts(testWebRequest{})
    if amount != 1 {
        t.Errorf("People in space, got: %d, want: %d.", amount, 1)
    }
}

I have an exported method called GetAstronauts which calls into a HTTP endpoint, reads the bytes from the result and then parses this into a struct and returns the integer in the "number" property.

My fake / test-double in the test only returns the bare minimum of JSON needed to satisfy the test, and to begin with I had it return a different number so that I knew the test worked. It's hard to be sure whether a test that passes first time has worked.

Here's the application code where we run our main function. The GetAstronauts function takes an interface as its first argument allowing us to isolate and abstract away any HTTP logic from this file and its import list.

package main

import (  
    "encoding/json"
    "fmt"
    "log"
)

func GetAstronauts(getWebRequest GetWebRequest) int {  
    url := "http://api.open-notify.org/astros.json"
    bodyBytes := getWebRequest.FetchBytes(url)
    peopleResult := people{}
    jsonErr := json.Unmarshal(bodyBytes, &peopleResult)
    if jsonErr != nil {
        log.Fatal(jsonErr)
    }
    return peopleResult.Number
}

func main() {  
    liveClient := LiveGetWebRequest{}
    number := GetAstronauts(liveClient)

    fmt.Println(number)
}

The GetWebRequest interface specifies the following function:

type GetWebRequest interface {  
    FetchBytes(url string) []byte
}

Interfaces are inferred on rather than explicitly decorated onto a struct. This is different from languages like C# or Java.

The complete file named types.go looks like this and was extracted from the previous blog post:

package main

import (  
    "io/ioutil"
    "log"
    "net/http"
    "time"
)

type people struct {  
    Number int `json:"number"`
}

type GetWebRequest interface {  
    FetchBytes(url string) []byte
}

type LiveGetWebRequest struct {  
}

func (LiveGetWebRequest) FetchBytes(url string) []byte {  
    spaceClient := http.Client{
        Timeout: time.Second * 2, // Maximum of 2 secs
    }

    req, err := http.NewRequest(http.MethodGet, url, nil)
    if err != nil {
        log.Fatal(err)
    }

    req.Header.Set("User-Agent", "spacecount-tutorial")

    res, getErr := spaceClient.Do(req)
    if getErr != nil {
        log.Fatal(getErr)
    }

    body, readErr := ioutil.ReadAll(res.Body)
    if readErr != nil {
        log.Fatal(readErr)
    }
    return body
}

Choosing what to abstract

The above unit test is effectively only testing the json.Unmarshal function and our assumptions about what a valid HTTP response body would look like. This abstracting may be OK for our example, but our code coverage score will be low.

It is also possible to do lower level testing to make sure that the HTTP get timeout of 2 seconds is correctly enforced, or that we created a GET request instead of a POST.

Fortunately Go has set of helper functions for creating fake HTTP servers and clients.

Going further:

  • explore the http/httptest package
  • and refactor the test above to use a fake HTTP client.
  • what is the test coverage percentage like before and after?

Questions and feedback

Please follow me on Twitter @alexellisuk to keep up to date with new content. You can also get in touch with questions, comments and suggestions.

See also:

Alex Ellis

Read more posts by this author.

United Kingdom http://alexellis.io/

Subscribe to alex ellis' blog

Get the latest posts delivered right to your inbox.

or subscribe via RSS with Feedly!