How to test a Go function which runs a command?

Question

Following the example at https://golang.org/pkg/os/exec/#Cmd.StdoutPipe, suppose I have a function getPerson() defined like so:

package stdoutexample

import (
    "encoding/json"
    "os/exec"
)

// Person represents a person
type Person struct {
    Name string
    Age  int
}

func getPerson() (Person, error) {
    person := Person{}
    cmd := exec.Command("echo", "-n", `{"Name": "Bob", "Age": 32}`)
    stdout, err := cmd.StdoutPipe()
    if err != nil {
        return person, err
    }
    if err := cmd.Start(); err != nil {
        return person, err
    }
    if err := json.NewDecoder(stdout).Decode(&person); err != nil {
        return person, err
    }
    if err := cmd.Wait(); err != nil {
        return person, err
    }
    return person, nil
}

In my 'real' application, the command run can have different outputs, I'd like to write test cases for each of these scenarios. However, I'm not sure how to go about this.

So far all I have is a test case for one case:

package stdoutexample

import (
    "testing"

    "github.com/stretchr/testify/assert"
    "github.com/stretchr/testify/require"
)

func TestGetPerson(t *testing.T) {
    person, err := getPerson()
    require.NoError(t, err)
    assert.Equal(t, person.Name, "Bob")
    assert.Equal(t, person.Age, 32)
}

Perhaps the way to go about this is to split this function into two parts, one which writes the output of the command to a string, and another which decodes the output of a string?

Answer 1

adding to https://stackoverflow.com/a/58107208/9353289,

Instead of writing separate Test functions for every test, I suggest you use a Table Driven Test approach instead. Here is an example,

func Test_getPerson(t *testing.T) {
    tests := []struct {
        name          string
        commandOutput []byte
        want          Person
    }{
        {
            name:          "Get Bob",
            commandOutput: []byte(`{"Name": "Bob", "Age": 32}`),
            want: Person{
                Name: "Bob",
                Age:  32,
            },
        },

        {
            name:          "Get Alice",
            commandOutput: []byte(`{"Name": "Alice", "Age": 25}`),
            want: Person{
                Name: "Alice",
                Age:  25,
            },
        },
    }
    for _, tt := range tests {
        t.Run(tt.name, func(t *testing.T) {
            got, err := getPerson(tt.commandOutput)
            require.NoError(t, err)
            assert.Equal(t, tt.want.Name, got.Name)
            assert.Equal(t, tt.want.Age, got.Age)

        })
    }
}

Simply adding test cases to the slice, will run all test cases.

Answer 2

your implementation design actively rejects fine grain testing because it does not allow any injection.

However, given the example, besides using a TestTable, there is not much to improve.

Now, on a real workload, you might encounter unacceptable slowdown dues to call to the external binary. This might justify another approach involving a design refactoring to mock and the setup of multiple tests stubs.

To mock your implementation you make use of interface capabilities. To stub your execution you create a mock that outputs stuff you want to check for.

package main

import (
    "encoding/json"
    "fmt"
    "os/exec"
)

type Person struct{}

type PersonProvider struct {
    Cmd outer
}

func (p PersonProvider) Get() (Person, error) {
    person := Person{}
    b, err := p.Cmd.Out()
    if err != nil {
        return person, err
    }
    err = json.Unmarshal(b, &person)
    return person, err
}

type outer interface{ Out() ([]byte, error) }

type echo struct {
    input string
}

func (e echo) Out() ([]byte, error) {
    cmd := exec.Command("echo", "-n", e.input)
    return cmd.Output()
}

type mockEcho struct {
    output []byte
    err    error
}

func (m mockEcho) Out() ([]byte, error) {
    return m.output, m.err
}

func main() {

    fmt.Println(PersonProvider{Cmd: echo{input: `{"Name": "Bob", "Age": 32}`}}.Get())
    fmt.Println(PersonProvider{Cmd: mockEcho{output: nil, err: fmt.Errorf("invalid json")}}.Get())

}

Answer 3

I added unit tests by splitting the function into two parts: one which reads the output to a slice of bytes, and one which parses that output to a Person:

package stdoutexample

import (
    "bytes"
    "encoding/json"
    "os/exec"
)

// Person represents a person
type Person struct {
    Name string
    Age  int
}

func getCommandOutput() ([]byte, error) {
    cmd := exec.Command("echo", "-n", `{"Name": "Bob", "Age": 32}`)
    return cmd.Output()
}

func getPerson(commandOutput []byte) (Person, error) {
    person := Person{}
    if err := json.NewDecoder(bytes.NewReader(commandOutput)).Decode(&person); err != nil {
        return person, err
    }
    return person, nil
}

The following test cases pass:

package stdoutexample

import (
    "testing"

    "github.com/stretchr/testify/assert"
    "github.com/stretchr/testify/require"
)

func TestGetPerson(t *testing.T) {
    commandOutput, err := getCommandOutput()
    require.NoError(t, err)
    person, err := getPerson(commandOutput)
    require.NoError(t, err)
    assert.Equal(t, person.Name, "Bob")
    assert.Equal(t, person.Age, 32)
}

func TestGetPersonBob(t *testing.T) {
    commandOutput := []byte(`{"Name": "Bob", "Age": 32}`)
    person, err := getPerson(commandOutput)
    require.NoError(t, err)
    assert.Equal(t, person.Name, "Bob")
    assert.Equal(t, person.Age, 32)
}

func TestGetPersonAlice(t *testing.T) {
    commandOutput := []byte(`{"Name": "Alice", "Age": 25}`)
    person, err := getPerson(commandOutput)
    require.NoError(t, err)
    assert.Equal(t, person.Name, "Alice")
    assert.Equal(t, person.Age, 25)
}

where the Bob and Alice test cases simulate different output which can be generated by the command.