Next: Extracting an ESM module from a Deno script
April 3, 2021

Getting started with Go

The Go programming language has become an important tool for developers, particularly around platforms like Kubernetes and Docker.

Go was created at Google by a team whose roots go back to Bell Labs and C. Their motivations included fast compilation time, and productive development of large scale distributed systems, handling high volumes of concurrent requests.

This article describes my experience as a new user of Go, building my first Go library. It follows a learning pattern similar to forays from Node to Rust.

Getting started

The Tour of Go is a great way to get familiar with the language syntax. I started with 'hello world' at, and found myself going back to the tour for different topics.

The macOS installer copies everything into /usr/local/go, so I opted to download the latest release from into a versioned $GOROOT under my home directory. Here's what I have in my '.bash_profile':

export GOPATH=~/go
export GOROOT=~/go1.16.3
export PATH=${PATH}:${GOROOT}/bin:${GOPATH}/bin

VS Code

The VS Code Go extension has improved a lot over the years. It now auto-installs the delve debugger, and the gopls language server. I did not have to do any additional configuration.

Hovering over types like Builder shows source docs, and links to

Porting from Rust to Go

I found it quite easy to port shortscale-rs to shortscale-go.

Go has no ownership syntax, and the run-time includes a garbage collector.

In this case, I was also lucky that the Go strings.Builder standard library package is very similar to the writer pattern I ended up using for Rust.

Overall, I was pleasantly surprised with the readability of the code

package shortscale

import (

// Shortscale converts numbers into English words.
// Supports positive integers from 0 to 999_999_999_999_999_999.
// Larger values return "(big number)".
func Shortscale(n uint64) string {
  if n <= 20 {
    return numwords[n]
  if n > 999_999_999_999_999_999 {
    return "(big number)"
  b := new(strings.Builder)
  writeScale(b, n, 1_000_000_000_000_000) // quadrillions
  writeScale(b, n, 1_000_000_000_000)     // trillions
  writeScale(b, n, 1_000_000_000)         // billions
  writeScale(b, n, 1_000_000)             // millions
  writeScale(b, n, 1_000)                 // thousands
  writeHundreds(b, n)
  writeTensAndUnits(b, n, b.Len() > 0)
  return b.String()

func writeTensAndUnits(b *strings.Builder, n uint64, ifAnd bool) {
  n = n % 100
  if n == 0 {
  if ifAnd {
    writeWord(b, "and")
  if n <= 20 {
    writeWord(b, numwords[n])
  writeWord(b, numwords[n/10*10]) // tens
  units := n % 10
  if units > 0 {
    writeWord(b, numwords[units])

func writeHundreds(b *strings.Builder, n uint64) {
  n = n / 100 % 10
  if n == 0 {
  writeWord(b, numwords[n])
  writeWord(b, numwords[100])

func writeScale(b *strings.Builder, n uint64, thousands uint64) {
  n = n / thousands % 1_000
  if n == 0 {
  writeHundreds(b, n)
  writeTensAndUnits(b, n, (n/100%10) > 0)
  writeWord(b, numwords[thousands])

func writeWord(b *strings.Builder, word string) {
  if b.Len() > 0 {
    b.WriteString(" ")

var numwords = map[uint64]string{
  0:                     "zero",
  1:                     "one",
  2:                     "two",
  3:                     "three",
  4:                     "four",
  5:                     "five",
  6:                     "six",
  7:                     "seven",
  8:                     "eight",
  9:                     "nine",
  10:                    "ten",
  11:                    "eleven",
  12:                    "twelve",
  13:                    "thirteen",
  14:                    "fourteen",
  15:                    "fifteen",
  16:                    "sixteen",
  17:                    "seventeen",
  18:                    "eighteen",
  19:                    "nineteen",
  20:                    "twenty",
  30:                    "thirty",
  40:                    "forty",
  50:                    "fifty",
  60:                    "sixty",
  70:                    "seventy",
  80:                    "eighty",
  90:                    "ninety",
  100:                   "hundred",
  1_000:                 "thousand",
  1_000_000:             "million",
  1_000_000_000:         "billion",
  1_000_000_000_000:     "trillion",
  1_000_000_000_000_000: "quadrillion",

Tests and benchmarks

The testing package provides support for running tests and benchmarks with go test. The GitHub Action workflow for shortscale-go make use of this.

Out of curiosity, I ran BenchmarkShortscale for two variants of the Shortscale function, one which pre-allocates memory for string.Builder, and one which does not. Pre-allocating, reduced the number of allocs/op from 4 to 1, improving ns/op by about 85ns.


$ go test -bench . -benchmem
goos: darwin
goarch: amd64
cpu: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz

5694252	       205.5 ns/op	      64 B/op	       1 allocs/op

Not pre-allocated

$ go test -bench . -benchmem
goos: darwin
goarch: amd64
cpu: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz

4100697	       292.9 ns/op	     120 B/op	       4 allocs/op

Dependency management

Until quite recently, Go did not have built-in package versioning like npm or cargo. This led to incompatibile versioning add-ons, like godep and glide, which made packages with nested dependencies difficult to consume. E.g. see this old from kubernetes/client-go.

Fortunately, Go modules are enabled as the default in Go since v1.16.


I created my shortscale-go module with go mod init following the guide in Using Go Modules.

$ go mod init
go: creating new go.mod: module

This created a new go.mod file with the following content.


go 1.16

I was a little surprised that there was no way to indicate the module version inside go.mod. Go relies on git tags in the form vx.x.x for this. As I pushed each version to GitHub, I used the GitHub Releases UI to create the tag.

The shortscale package is published at

It turns out that fetching any versioned module with go get, automatically adds that module to the registry at This feels a little strange at first, but the more I use it, the more I think it's a clever solution.

How about using a similar scheme to create a vendor-neutral registry for ESM modules?

To leave a comment
please visit

(c) Jürgen Leschner