Skip to content


Folders and files

Last commit message
Last commit date

Latest commit



43 Commits

Repository files navigation

This document assumes Go version 1.0.3.

Table of contents

The go tool is bundled with Go distribution by default and it's convenient for automating common tasks such as getting dependencies, building, and testing your code. It's easy to use and provides a consistent command-line interface and it also expects you to respect a bunch of conventions, some of which I find peculiar and think they introduce a slight learning curve for some users and require a bit of getting used to.

While the conventions imposed by go tool might seem natural for a hardcore gopher, it takes effort for a newcomer to get up to speed with it. If you hit a wall trying to make it work for you and ask for help on #go-nuts channel or golang-nuts group showing your code layout and error messages go get or go build produces, you will most likely be told to first learn how to use go tool properly before you start coding in Go. Its documentation is actually pretty good, but it's not easy to absorb it all at once.

My experience was such that neither the recommended initial reading, nor discussions on the mailing list cleared up the picture completely for me. I was only able to eventually grasp the go way by gathering tidbits from the net, through experimentation, and by looking at the go tool's source code.

In this article I'm going to explain the go way from an outsider's point of view. Assuming you're likely to encounter similar hurdles along your way, this guide should answer your questions and help you understand go tool's conventions. There is also a FAQ with code samples at the bottom.

Here are some fundamentals you need to be aware of when using go tool.

1. Go tool is only compatible with code that resides in a workspace

This is the fundamental rule of writing Go code. There are some exceptions that apply to little programs and throw-away code, but in general everything you write in Go will reside in your workspace. So what is a workspace?

A workspace is a directory in your file system, the path to which is stored in the environment variable GOPATH. This directory has a certain structure, in the most basic terms it should have one subdirectory named src. Within the latter you create new subdirectories, one for each separate Go package.

This is described in more detail in the aforementioned article and in the FAQ near the end of the present document. To keep it simple, always create a new directory inside $GOPATH/src when you're starting a new Go project.

2. Go tool does not allow you to depend on specific versions of external packages

If you really need to have a specific version of a certain package, you'll need to fork that package and check it out at the specific version you desire. If you need to use different versions of a single package, you'll need to create a separate fork for each version.

Obviously, this approach doesn't scale and gets tedious once you need more than one version of any given package. Go's reasoning behind this is that versioning is so damn hard that your dependencies should better all be working for you at their latest versions and also at the latest versions of their respective dependencies, and so on recursively. While this is a sound advice, it is generally not possible to adhere to it in practice. You will need to depend on specific versions of your immediate dependencies or dependencies of your dependencies, etc.

In a nutshell, don't try to emulate versioning with go tool, it just doesn't play well with it.

3. When working with Go tool, use fully qualified imports and always build import paths from $GOPATH

There is no such thing as local packages in Go. While local imports are possible to some extent, they're meant more for the Go devs themselves than for Go users.

Anything you import is relative to your $GOPATH/src. Thus, if you have a directory structure like the following one:

└── src
    └── gopher
        ├── main.go
        └── sub
            └── sub.go

in your main.go file you'll need to import sub as follows:

package main

import "gopher/sub"

func main() {

If you'd like to use a go package that's publicly available from some code hosting site, the recommended approach is to import it like this:

import ""

func main() {

Go tool let's you download remote packages either by passing their import path to go get or calling go get ./... inside your project's directory to recursively get all remote dependencies.

go get

The source for the downloaded package will end up in $GOPATH/src/ Go tool will also automatically build a static lib and put it in $GOPATH/pkg/.... To read more about this, run go help get and go help gopath. Also, see question 10.

Coming from other languages/environments, you may expect that go is a full package management solution. It isn't! The following FAQ entries might be useful:

The best way to go about GOPATH for a beginner is to set it and forget it. If you're just starting with Go, do this:

  1. Install Go.

  2. Choose a directory where you'll be keeping all of your Go code and set GOPATH to it, e.g.

    # in your .bashrc or similar file
    export GOPATH=$HOME/go
  3. Forget about it. Start every new project inside your $GOPATH/src and use the go tool for building, testing, fetching dependencies.

If you'd like to know more about GOPATH, it's frequently mentioned throughout this FAQ. Also, take a look at the following resources.

Before you start any coding, you should pick a directory that will become your Go workspace. All your Go code will reside there. Set the GOPATH environment variable to the path to that directory in your .bashrc or similar file for the shell you're using.

export GOPATH=/Users/alco/go

You'll also need to create a subdirectory named src inside your $GOPATH, this is where you'll be keeping your Go packages and commands. Here's what your initial directory structure is going to look like:

$ tree -L 2 $GOPATH
└── src
    ├── example
    ├── gopher
    └── testy

Each of the subdirectories inside src represents a separate package or a command (see question 11 describing the difference between packages and commands). Each one will contain at least one .go file and may also have subdirectories of its own.

For more information about GOPATH and workspace directory structure, run go help gopath.

Navigate to your package's directory and use the go tool to build and run your code.

Let's assume you created a program in $GOPATH/src/example directory:

$ cd $GOPATH/src/example
$ ls

$ cat main.go
package main

import "fmt"

func main() {
    fmt.Println("Hello world!")

The quickest way to run it is using go run:

$ go run main.go
Hello world!

If your main package is split into multiple files (see question 3 for details on how to do that), you will need to pass them all as arguments to go run.

Soon, however, you'll want to produce a binary from your Go source that you can run as a standalone executable, without using go tool. Use go build for that, it will create an executable in the current directory.

You can also run go install, it will build the code and place the executable in $GOPATH/bin. You might want to add the latter to your PATH environment variable to be able to run your Go programs anywhere in the file system. If you set GOBIN environment variable, then results of running go install will be placed there.

$ go build
$ ls
example main.go

$ ./example
Hello world!

$ go install
$ $GOPATH/bin/example
Hello world!

Here we defined a main package which has a main function that is the starting point of our Go program. Any Go program must have exactly one main function. By convention, these packages (the ones that declare package main) are called commands. In other words, we have built a command named example.

You can also define packages with other names, those are called simply packages. They are not intended to produce executable programs, but rather be included as part of some command that will provide a main package with a single main function defined in it.

See also: go help run, go help build, go help install.

Go treats files in a single directory as belonging to one package as long as they all have the same name in their package declarations.

Let's continue working on our example command. We'll add a second file named helper.go and define a helper function in it.

$ cd $GOPATH/src/example
$ cat helper.go
package main

func privateHelperFunc() int {
    return 25

# now edit main.go to call this function
$ tail -3 main.go
func main() {
    fmt.Println("Hello world! My lucky number is", privateHelperFunc())

Now we cannot simply go run main.go because main.go calls a function defined in another file. We can either pass all files as arguments to go run or build the current package and then run the produced binary.

$ go run main.go
# command-line-arguments
./main.go:6: undefined: privateHelperFunc

$ go run *.go
Hello world! My lucky number is 25

$ go build
$ ./example
Hello world! My lucky number is 25

Private (non-exported) functions and data are accessible in all files that belong to a single package.

A main package allows only one main function to be defined, so you'll need to choose one single file from your main package to put it in.

Finally, you can also split any other package into multiple files, not just the main ones. Go's standard packages use this ability pervasively.

Subpackages are just separate packages that happen to reside in another package's directory. Go doesn't treat them in any special way, so import paths for subpackages are relative to your $GOPATH/src. Use a subpackage only when its functionality is tied to the main package which contains it and when it doesn't make sense to put that package on one level with other top-level packages.

Let's create a subdirectory in our example project called math and create a file there named math.go.

$ cat math/math.go
package math

func Mul2(x int) int {
    return x * 2

Let's also edit main.go to call Mul2().

$ cat main.go
package main

import (
    "example/math"  // just "math" would not work, it would import std package math

func main() {
    fmt.Println("Hello world! My lucky number is", math.Mul2(privateHelperFunc()))

$ go run *.go
Hello world! My lucky number is 50

Note that if you were to replace package math in math.go with package main, this would not make math.go part of the same main package that both main.go and helper.go belong to. It would be treated as another main package and you would get an error when trying to run or import it because it doesn't define a main function.

In the previous questions we were mainly looking at writing so called commands — packages that declare package main and are meant to be built to produce an executable binary.

The other flavor of Go packages is used as libraries or modules in other languages, you can't build them into an executable. Their purpose is to be imported into another package (not necessarily main package) to provide useful functionality to that package.

You create a package the same way as you would create a command. The only difference is that instead of package main you write package <some other name>. All other rules described in previous questions apply to these packages as well: you may split one package into multiple files, but all files belonging to a package reside in a single directory.

Let's say we have created a package called util that resides in $GOPATH/src/util.

$ cd $GOPATH/src/util
$ cat main.go   # file name is arbitrary and doesn't make any significance to Go
package util

import "math"

func Square(x float32) float32 {
    return x * x

func Circle(r float32) float32 {
    return math.Pi * r * r

func cube(x float32) float32 {
    return x * x * x

$ go build
# no output

In the context of a simple (non-main) package, go build is used to verify that the package compiles without errors. It doesn't produce any binary. To precompile the package for use by other package, you can run go install. This will put util.a into $GOPATH/pkg/<arch>/util.a. Read more about the pkg directory by running go help gopath.

In our util package, there are two exported functions (Square and Circle) and one private function (cube). The private function is only visible in files that are part of the package. Other packages can only call exported functions.

See also this wiki page, it describes the process of publishing and using remote packages.

Those two behave a bit differently depending on whether you're building a command (a main package) or a simple package.

When building a main package, the resulting executable will be placed in the current directory. Running go install, on the other hand, will build the source as usual and place the executable in $GOPATH/bin (unless you have GOBIN environment variable set, in which case resulting binary will be placed there). This has also been discussed in the second question.

Running go build inside the directory of a simple package does not produce any binary. Building a package this way is used to verify that it compiles cleanly. In order to create a binary from package source, run go install inside the package's directory. As discussed in the previous question, this will create an .a file inside $GOPATH/pkg/<arch> directory. Go tool will then be able to pick this binary when building other packages that import the current one.

See also: go help build, go help gopath.

Short answer — you don't. Go tool does expect you to work in a single workspace. You can add more than one to your GOPATH environment variable, but there's a gotcha: go get will always download new packages into the first location listed in GOPATH.

So while adding more than one path to GOPATH is rarely useful, you might still want to use multiple workspaces. You'll just need to change your GOPATH according to the workspace you're currently working in. There's definitely room here for some build tool that automates the bookkeeping.

See also question 14 for one example of using multiple workspaces.

No. You can change your GOPATH though, as described in the previous answer. But keep in mind that $GOPATH points to a workspace. Your packages go into the src directory inside a workspace.

If you're not going to import anything outside of standard library or have one level of local imports, then it'll work for you with go tool. If, however, you need to use fully qualified imports, you have to move your code to a workspace or else you'll get problems when trying to go get your dependencies and other bad things might also happen.

Workarounds are possible for particular cases and those can be provided by another tool. In general, however, you have to stick to Go's conventions to make it work for you.

To get all dependencies for the current package:

go get ./...

To download a particular remote package:

go get <package import path>  # see 'go help packages' for details

# for instance,
go get

All downloaded packages end up in $GOPATH/src. They are also automatically built and installed into $GOPATH/pkg. You can skip installation by passing -d flag to go get:

go get -d

Both kinds of packages live side by side in your workspace's src directory, so there's no distinction at the file system level. There is a convention to call the former ones simply packages and the latter ones commands. So, if your package's first line reads package main, it's a command. Otherwise, it's just called a package.

I'd recommend against that. Using multiple workspaces is tricky, you'll need to remember to adjust GOPATH every time you switch between them. See also question 7 for more details.

Sure, but you'll need to provide an alias during import so that the package's main function does not collide with your main function, if you're importing it into a main package.

import chef ""

Don't let it bother you. This is a practice already widely adopted by the Go community. It also works well with go get automatically fetching or updating your dependencies. Stick to it, and you won't have troubles working with other people's packages and sharing packages of your own.

If you're still curious, you can actually move stuff around inside $GOPATH/src after go get has downloaded your dependencies. This would go against the Go way though. Also, currently, the only fast way to distinguish other people's packages from your own is by looking at the directory structure inside your $GOPATH/src: remote packages will reside in one of the directories like or So it's handy to leave that way.

Keeping remote packages in another directory, separate from your own packages, would certainly be great. And it can be emulated by specifying more than one path in GOPATH. Running go get will always download packages into the first directory listed in GOPATH, so you can then create your own packages in the directory which is second on the list of paths in GOPATH.

Using multiple workspaces has its issues though. See question 7 for more details.

There is no widely adopted solution for package version management, currently. The go tool has no knowledge of package versions; though it does choose package versions (vcs tags) based on your current go version.

The currently accepted approach is to build all of your dependencies off of master. Running go get -u ./... to update your locally cached dependencies is helpful to stay out of trouble, though not ideal.

Before you get up in arms and race off to write your own package version manager, be sure to read through discussion of the previous attempts at this:

go tool does not provide any way to create a reproducible environment for fool-proof deployments.

If you tested your code locally, you can never be sure that it'll work during your next deploy, because one of the dependencies might introduce a breaking change during the time period between your testing and deployment.

Use dep to manage and lock your dependencies. It's not a part of the toolchain yet, but going to be included in the future releases.

Take a look at this list of official Go talks. Those are most helpful to start getting into the mindset of a Go programmer. Wish they were mentioned more prominently on the official Go site.

See this wiki page.


A getting started guide for Go newcomers






No releases published


No packages published