~
Imagine your project currently have 0% unit test code coverage. And your boss keep pushing it to 80% or even 90%? What do you do? Give up?
What if I tell you there is a way? Not entirely cheating but ... you know, there is always trade off.
If your purpose is to test carefully all path, check if all return is correctly. Sadly this post is not for you, I guess. If you only want good number on test coverage, with minimum effort as possible, I hope this will show you some idea you can use :)
In my opinion, unit test is not that important (like must must have). It's just make sure your code is running excatly as you intent it to be. If you don't think about edge case before, unit test won't help you.
When I learn programming, I encounter very interesting idea, which become mainly my mindset when I dev later. I don't recall it clearly, kinda like: "Don't just fix bugs, rewrite it so that kind of bugs will not appear again". So in our context, there is some thing we hardly or can not write test in Go. My suggestion is don't use that thing.
In my experience, I can list a few here:
viper.Get...
). You can and
you should init all config when project starts.
Err...
). You should move
all global var to fields inside some struct.
If you code Go long enough, you know table driven tests and how is that so useful. You set up test data, then you test. Somewhere in the future, you change the func, then you need to update test data, then you good!
In simple case, your func only have 2 or 3 inputs so table drive tests is still looking good. But real world is ugly (maybe not, idk I'm just too young in this industry). Your func can have 5 or 10 inputs, also your func call many third party services.
Imagine having below func to upload image:
type service struct {
db DB
redis Redis
minio MinIO
logService LogService
verifyService VerifyService
}
func (s *service) Upload(ctx context.Context, req Request) error {
// I simplify by omitting the response, only care error for now
if err := s.verifyService.Verify(req); err != nil {
return err
}
if err := s.minio.Put(req); err != nil {
return err
}
if err := s.redis.Set(req); err != nil {
return err
}
if err := s.db.Save(req); err != nil {
return err
}
if err := s.logService.Save(req); err != nil {
return err
}
return nil
}
With table driven test and thanks to stretchr/testify, I usually write like this:
type ServiceSuite struct {
suite.Suite
db DBMock
redis RedisMock
minio MinIOMock
logService LogServiceMock
verifyService VerifyServiceMock
s service
}
func (s *ServiceSuite) SetupTest() {
// Init mock
// Init service
}
func (s *ServiceSuite) TestUpload() {
tests := []struct{
name string
req Request
verifyErr error
minioErr error
redisErr error
dbErr error
logErr error
wantErr error
}{
{
// Init test case
}
}
for _, tc := range tests {
s.Run(tc.name, func(){
// Mock all error depends on test case
if tc.verifyErr != nil {
s.verifyService.MockVerify().Return(tc.verifyErr)
}
// ...
gotErr := s.service.Upload(tc.req)
s.Equal(wantErr, gotErr)
})
}
}
Looks good right? Be careful with this. It can go from 0 to 100 ugly real quick.
What if req is a struct with many fields? So in each test case you need to set up req. They are almost the same, but with some error case you must alter req. It's easy to be init with wrong value here (typing maybe ?). Also all req looks similiar, kinda duplicated.
tests := []struct{
name string
req Request
verifyErr error
minioErr error
redisErr error
dbErr error
logErr error
wantErr error
}{
{
req: Request {
a: "a",
b: {
c: "c",
d: {
"e": e
}
}
}
// Other fieles
},
{
req: Request {
a: "a",
b: {
c: "c",
d: {
"e": e
}
}
}
// Other fieles
},
{
req: Request {
a: "a",
b: {
c: "c",
d: {
"e": e
}
}
}
// Other fieles
}
}
What if dependencies of service keep growing? More mock error to test data of course.
tests := []struct{
name string
req Request
verifyErr error
minioErr error
redisErr error
dbErr error
logErr error
wantErr error
// Murr error
aErr error
bErr error
cErr error
// ...
}{
{
// Init test case
}
}
The test file keep growing longer and longer until I feel sick about it.
See
tektoncd/pipeline unit test
to get a feeling about this. When I see it, TestPodBuild
has
almost 2000 lines.
The solution I propose here is simple (absolutely not perfect, but good with my usecase) thanks to stretchr/testify. I init all default action on success case. Then I alter request or mock error for unit test to hit on other case. Remember if unit test is hit, code coverage is surely increased, and that my goal.
// Init ServiceSuite as above
func (s *ServiceSuite) TestUpload() {
// Init success request
req := Request{
// ...
}
// Init success action
s.verifyService.MockVerify().Return(nil)
// ...
gotErr := s.service.Upload(tc.req)
s.NoError(gotErr)
s.Run("failed", func(){
// Alter failed request from default
req := Request{
// ...
}
gotErr := s.service.Upload(tc.req)
s.Error(gotErr)
})
s.Run("another failed", func(){
// Alter verify return
s.verifyService.MockVerify().Return(someErr)
gotErr := s.service.Upload(tc.req)
s.Error(gotErr)
})
// ...
}
If you think this is not quick enough, just ignore the response. You only need to check error or not if you want code coverage only.
So if request change fields or more dependencies, I need to update success case, and maybe add corresponding error case if need.
Same idea but still with table, you can find here Functional table-driven tests in Go - Fatih Arslan.