Christoph Hochstrasser

Go Patterns

I must admit that I’m a serious fan of Go. I spent most of the last company hackdays with it, doing some small projects, like a simple language independent job queue, and some smaller libraries.

After reading over my code, I noticed that there is code I’m writing over and over again.

Many of these patterns are from Effective Go, a book which is maintained by the Go team. This book is available online, and highly recommended.

If you are not already familiar with Go, I also recommend going through the Tour of Go to get to know the core concepts of the language before reading this article.

Constructor Function

Constructor functions are functions which return an initialized reference to the type. They have the prefix New (or new when private). This means when there is a type T then the name of the constructor function is NewT.

A package doesn’t need to provide a constructor function for every type. Usually constructor functions are only needed when the type needs initialization, i.e. when the type embeds references to other types, like Maps or references to other Structs.


type Event struct {}

type EventEmitter struct {
    Events map[string][]func(*Event)

func NewEventEmitter() *EventEmitter {
    return &EventEmitter{
        Events: make(map[string][]func(*Event)),


I’m not going to explain in detail what Futures are, because it’s not in the scope of this article. TL;DR: If you have an operation which usually takes rather long (e.g. sending on a socket), then you give the caller a “Future” immediately. When the operation is finished the Future gets resolved, which notifies the caller.

In Go you can replicate most of the behavior of Futures just with a couple lines of code — thanks to channels and Goroutines.

I like to write Futures in Go with a struct type which has Success and Error channels. These channels are used to resolve the Future into the “error” and “success” states, respectively.

// Can be anything
type Result interface{}

type MyFuture struct {
    Success chan Result
    Error chan error

func doSomethingAsync() *MyFuture {
    future := &MyFuture{Success: make(chan Result), Error: make(chan error)}

    go func() {
        time.Sleep(5 * time.Second)
        future.Success <- 1

    return future

The “Then” part uses a select construct to wait on the Success and Error channels until the Future is resolved.

future := doSomethingAsync()

select {
case result := <-future.Success:
    // Future was successfully resolved and we can work with the
    // result
case err := <- future.Error:
    // Future was resolved as error

If you read throughly about Futures in Programming, then you might notice that this implementation is fairly incomplete. It lacks stacking of futures, error bubbling and protection against resolving from the outside. Constructing a more “correct” Future implementation though is unnecessary for most use cases, and left as an exercise to the reader.


Just like C has “BYOB” — Bring your own Buffer, Go has “BYOC” — __B__ring __Y__our __O__wn __C__hannel.

This means that functions which produce or consume values with a channel let the user make the channel.

The advantage is that users are able to decide the buffer size of the channel. This is helpful when you want to run multiple instances of the function on the same channel.

package main

import "fmt"
import "math/rand"

// Generates some random numbers
func generateSomeRandoms(c chan int) {
    for {
        number := rand.Int()


func main() {
    numWorkers := 4
    c := make(chan int, numWorkers)

    for i := 0; i < numWorkers; i++ {
        go generateSomeRandoms(c)

    // Get 10 random numbers
    for i := 0; i < 10; i++ {



Generators are a feature known from Python or more recently also from PHP and JavaScript. Generators represent non-rewindable sequences of values. In Go we can represent a generator by a function which yields values on a channel via an embedded Goroutine.

package main

import "fmt"

func xrange(from int, to int) chan int {
    yield := make(chan int)

    go func() {
        for i := from; i <= to; i++ {
            yield <- i

    return yield

func main() {
    for i := range xrange(0, 10) {

I think it’s fine that generators don’t adhere to BYOC, because it usually doesn’t make sense to run multiple instances of the same generator on the same channel.

Generators are a good choice for offering an efficient and uniform iteration API for custom data structures, like Stacks, Lists or Trees. The range construct is able to iterate over a channel, and while the for-range loop processes the current value, the generator is able to already calculate the next value.


Go is a simple and effective language with powerful primitives, which can be used to replicate many features of higher level languages.

Especially Channels and Goroutines are extremely powerful, yet simple and efficient primitives. I really love how Go makes Concurrency really easy, without the bloat, and in a safe way.