Skip to end of metadata
Go to start of metadata

These guidelines for writing Go code are WIP

Overview

This document covers common coding styles and guidelines for all ForgeRock products.

Copyright notices

Within the FRaaS codebases we are not currently adding licence headers to individual source files.  This practice diverges from the standard practice of doing so across other projects.

The ForgeRock Go coding style

Quick dos/donts

  • Don't use os.Exit (or logging.Fatal) anywhere except the main function - this stops any 'defer' statements from running and also prints no message when it exits, which can cause confusion when trying to debug why a program suddenly stopped. If something has failed in such a way that it can't recover, panic instead. If you can't do anything with the error, just return the error from the function you are in (and handle this case further up).
  • If you are returning an error from another package (or a standard library function), wrap it with errors.Wrap from github.com/pkg/errors which will add context to the error to make it easier to understand where it comes from. If you are creating a new error, use errors.New (or errors.Errorf instead of fmt.Errorf) from the same library which provide the same utility.

Documentation

  • Each Go project should include one or more README.md files detailing
    • An overview of the project and its purpose
    • How to build, test, deploy and configure the executable
  • All public constants, structs, fields, interfaces and functions must have Go doc
  • All packages must have Go doc - Packages with more than one source file should consider providing package documentation using a doc.go file

Source code layout

In addition to good documentation, having a consistent approach to organising code across directories and within a given source file makes it easier for engineers to move between projects and get up to speed quickly.

  • Each Go project should use a standard layout
  • Each Go source file should, as far as possible, be readable top-to-bottom with public / high level functions at the top of source files and private / helper functions lower down:
    • vars/consts
    • interfaces
    • structs
    • methods
    • constructors
    • public functions
    • private functions

Linting rules

Where possible, agreed standards relating to source files should be enforced by linting during continuous integration.  The linting rules currently in use by the FRaaS team are:

golangci-lint_config_base.yml
linters:
  # inverted configuration with `enable-all` and `disable` is not scalable during updates of golangci-lint
  disable-all: true
  enable:
    - deadcode
    - errcheck
    - gofmt
    - goimports
    - gosimple
    - govet
    - ineffassign
    - structcheck
    - typecheck
    - unused
    - varcheck

run:
  tests: true

Logging

  • Avoid logging directly at the site where an error is produced or returned.  Instead let the entry point to processing do the logging. If the returned error message is insufficient (often they are already sufficient) use errors.Wrap to add context to the returned error.
  • Use the github.com/pkg/errors package instead of errors, and user errors.WithStack(err) wherever an error is produced or returned from an external package. errors.WithStack produces a stack trace pointing to the line of code which produced the error, which also prevents us from having to add our own custom error message so that we can correlate the error message to the line of code which produced it.This can also be done wherever there's an errors.New, I.E errors.WithStack(errors.New("my error")).
  • Any new packages that need to log things should use the common logger specified in common/pkg/logging/logging.go. This can be used by importing in your package (in package.go, or another file)

    go log example
    package newpackage
    
    import (
    	"github.com/ForgeCloud/saas/go/common/pkg/logging"
    )
    
    var (
    	log = logging.Recorder
    )
  • Programs should only exit in a 'main' function, and should log which error caused the program to fail at the 'Panic' level. This will make it obvious when looking at logs where and why the program exited.

    go error handling example
    func main() {
    	// ... do some setup ...
    	err := runProgramForReal()
    	if err != nil {
    		log.WithError(err).Panic("Error while running program")
    		// or
    		// panic(err)
    	}
    }

Idiomatic Go

In addition to the points raised above, we should endeavour to write idiomatic Go.  Guidance for what these idioms are and how to follow them can be found in:

Mocking

Mocking is done using https://github.com/vektra/mockery which generates mocks using https://github.com/stretchr/testify.

What not to do
  • Do not write a 'custom' mock - eg, write a struct which implements the interface you want to mock. If somebody needs to change the interface further down the line, they will then need to modify your mock as well. When using an autogenerated mock, they can just run 'go generate' to recreate it.
  • Do not use a mock for an interface in the same package. Mocks should provide a description of the interface for something in the package - the package itself should not need to test the mock.
Generating mocks

When adding a new interface that will require a mock there are a few simple steps to follow

  1. Follow the instructions to install mockery
  2. Add a line either above the interface you've added or in a package.go file alongside the interface with a 'go generate' comment
  3. Run 'go generate ./...' in the relevant folder - this will run 'mockery' and generate a mock for your interface under the 'mocks' folder

Your interface should look like this:

go mocking example
//go:generate mockery --all

// ThingDoer does a thing
type ThingDoer interface {
	DoThing() (error)
}

If you need to generate a mock for an interface but you also need to use that mock in the same package, this will cause import cycles. To get around this, add '--inpackage' to the list of arguments to 'mockery' in the generate comment.

Using mocks

If you want to use one of these mocks in one of your tests, there is a small utility function in go/common/pkg/testutil/mockhelper.go which can be used to do some common setup and mock assertion. An example of how to use this:


go mock example
func TestSomething(t *testing.T) {
	// create mock controller
	ctrl := testutil.NewController(t)
	// defer 'finish' - this will check that your mock assertions are satisfied
	defer ctrl.Finish()

	// Create the mock object
	mockThingdoer := &mocks.Thingdoer{}

	// register the mock object with the controller - this will prevent mock assertions from instantly panicking, and will check for assertions at the end of the test
	ctrl.Register(&mockThingdoer.Mock)

	// Set mock assertions
	mockThingDoer.On("DoThing").Return(nil)

	// call your function, do basic checks
	err := useThingdoer(mockThingdoer)
	assert.NoError(t, err)

	// If 'DoThing' was not called, mockery will fail the test when the function ends
}


The testify documentation will have more information about what methods these mocks will have and how to use them.

Assertions in tests

We use https://github.com/stretchr/testify for writing assertions in tests. This has 2 common operations

  • 'assert' functions will fail the test if the assertion is incorrect, but the test will continue to run.
  • 'require' functions will fail the test and immediately stop the test if the assertion is incorrect.

'require' should be used when it's impossible to continue the test, such as if an unrecoverable error happens during test setup or if a particular object used is not valid:

object, err := GetObject()
require.NoError(t, err)
require.NotNil(t, object)

object.DoThing()

If a test can continue after something fails, use 'assert' instead.

Adding 'options' to a constructor

This is a fairly standard pattern to add extra options to a constructor for a struct. Say you have a struct with a constructor like this

type Recipe struct {
	name       string
	hasSpinach bool
}

func NewRecipe(name string, hasSpinach bool) Recipe {
	return Recipe{
		name:       name,
		hasSpinach: hasSpinach,
	}
}

If you want to extend this struct, the obvious way to do it is to add extra things to the constructor:

type Recipe struct {
	name       string
	hasSpinach bool
    numEggs    int
}

func NewRecipe(name string, hasSpinach bool, numEggs int) Recipe {
	return Recipe{
		name:       name,
		hasSpinach: hasSpinach,
        numEggs:    numEggs,
	}
}

In future people might want to add even more things to this structure, and this approach does not scale:

  • People might not care about any other fields in the struct other than the ones they care about
  • Every time people instantiate a struct, they need to pass a large number of arguments
  • Every time this signature changes, every other place in the code that uses this constructor needs changing as well

If you find yourself adding a lot of fields to a struct like this, use 'options' instead.

  • Set sensible defaults for the fields in the contructor
  • Specify a private interface controlling how to modify these fields (so that people can't define their own options, to keep the implementation in one package)
  • Add any options as types which implement this interface
type Recipe struct {
	name       string
	hasSpinach bool
    numEggs    int
}

type option interface {
	apply(*Recipe)
}

type WithEggs int
func (o WithEggs) Apply(r *Recipe) {
	r.numEggs = o
}

type WithSpinach bool
func (o WithSpinach) Apply(r *Recipe) {
	r.hasSpinach = o
}

func NewRecipe(name string, options ...option) Recipe {
	r := &Recipe{
		name:       name,
	}

	for _, o := range options {
		o.apply(r)
	}

	return r
}

Now anybody who wants to create one of these structs can pass options defining only what they care about rather than every single field in the struct.

recipe1 := NewRecipe("something")
recipe2 := NewRecipe("omelette", WithSpinach(true), WithEggs(6))

Sub-tests

If you are writing a test where there are multiple 'sub tests' (such as checking the behaviour of a certain endpoint with multiple different data inputs), Use t.Run() to separate out these sub-tests.

If you have a test setup like this:

func TestFunction(t *testing.T) {
	tests := []struct {
		name string
		input string
	}{
		{
			...
		},
	}

	for _, test := range tests {
		err := Function(test.input)
		require.NoError(t, err)
	}
}

Then as soon as any test runs into an error, it will implicitly fail all other tests as well. Instead, use sub-tests like this:

	for _, test := range tests {
		t.Run("test.name", func(t *testing.T) {
			err := Function(test.input)
			require.NoError(t, err)
		})
	}

Creating GCP API clients

When creating an API client (for example, a logging client that is used to read logs from the GCP API) and you only ever want to read data, pass a 'WithScope' option to the constructor to limit what the client can do:

crmService, err := cloudresourcemanager.NewService(ctx, option.WithScopes(cloudresourcemanager.CloudPlatformReadOnlyScope))
if err != nil {
	return errors.Wrap(err, "creating service")
}

...


Commonly used libraries

  • No labels

2 Comments

  1. Might be worth adding something about mock folders as well

  2. Unknown User (summer.mousa)

    Ginkgo is a popular BDD testing library for go, and has a richer command set than the builtin go testing interface.  (https://github.com/onsi/ginkgo)

    Also, this ticket was open because a lot of our go packages wasn't probably setting up tests: FRAAS-7867 - Getting issue details... STATUS