xesite/blog/within-go-repo-layout-2020-09-07.markdown

title	date	series	tags
The Within Go Repo Layout	2020-09-07	howto	go standards

The Within Go Repo Layout

Go repository layout is a very different thing compared to other languages. There's a lot of conflicting opinions and little firm guidance to help steer people along a path to more maintainable code. This is a collection of guidelines that help to facilitate understandable and idiomatic Go.

At a high level the following principles should be followed:

• If the code is designed to be consumed by other random people using that repository, it is made available for others to import
• If the code is NOT designed to be consumed by other random people using that repository, it is NOT made available for others to import
• Code should be as close to where it's used as possible
• Documentation helps understand why, not how
• More people can reuse your code than you think

Folder Structure

At a minimum, the following folders should be present in the repository:

• cmd/ -> houses executable commands
• docs/ -> houses human readable documentation
• internal/ -> houses code not intended to be used by others
• scripts/ -> houses any scripts needed for meta-operations

Any additional code can be placed anywhere in the repo as long as it makes sense. More on this later in the document.

Additional Code

If there is code that should be available for other people outside of this project to use, it is better to make it a publicly available (not internal) package. If the code is also used across multiple parts of your program or is only intended for outside use, it should be in the repository root. If not, it should be as close to where it is used as makes sense. Consider this directory layout:

repo-root
├── cmd
│   ├── paperwork
│   │   ├── create
│   │   │   └── create.go
│   │   └── main.go
│   ├── hospital
│   │   ├── internal
│   │   │   └── operate.go
│   │   └── main.go
│   └── integrator
│       ├── integrate.go
│       └── main.go
├── internal
│   └── log_manipulate.go
└── web
    ├── error.go
    └── instrument.go

This would expose packages repo-root/web and repo-root/cmd/paperwork/create to be consumed by outside users. This would allow reuse of the error handling in package web, but it would not allow reuse of whatever manipulation is done to logging in package repo-root/internal.

repo-root/cmd/

This folder has subfolders with go files in them. Each of these subfolders is one command binary. The entrypoint of each command should be main.go so that it is easy to identify in a directory listing. This follows how the go standard library does this.

For example:

repo-root
└── cmd
    ├── paperwork
    │   └── main.go
    ├── hospital
    │   └── main.go
    └── integrator
        └── main.go

This would be for three commands named paperwork, hospital, and integrate respectively.

As your commands get more complicated, it's tempting to create packages in repo-root/internal/ to implement them. This is probably a bad idea. It's better to create the packages in the same folder as the command, or optionally in its internal package. Consider if paperwork has a command named create, hospital has a command named operate and integrator has a command named integrate:

repo-root
└── cmd
    ├── paperwork
    │   ├── create
    │   │   └── create.go
    │   └── main.go
    ├── hospital
    │   ├── internal
    │   │   └── operate.go
    │   └── main.go
    └── integrator
        ├── integrate.go
        └── main.go

Each of these commands has the logic separated into different packages.

paperwork has the create command as a subpackage, meaning that other parts of the application can consume that code if they need to.

hospital has the operate command inside its internal package, meaning only cmd/foo/ and anything that has the same import path prefix can use that code. This makes it easier to isolate the code so that other parts of the repo cannot use it.

integrator has the integrate command as a separate go file in the main package of the command. This makes the integrate command code only usable within the command because main packages cannot be imported by other packages.

Each of these methods makes sense in some contexts and not in others. Real-world usage will probably see a mix of these depending on what makes sense.

repo-root/docs/

This folder has human-readable documentation files. These files are intended to help humans understand how to use the program or reasons why the program was put together the way it was. This documentation should be in the language most common to the team of people developing the software.

The structure inside this folder is going to be very organic, so it is not entirely defined here.

repo-root/internal/

The internal folder should house code that others shouldn't consume. This can be for many reasons. Generally if you cannot see a use for this code outside the context of the program you are developing, but it needs to be used across multiple packages in different areas of the repo, it should default to going here.

If the code is safe for public consumption, it should go elsewhere.

repo-root/scripts/

The scripts folder should contain each script that is needed for various operations. This could be for running fully automated tests in a docker container or packaging the program for distribution. These files should be documented as makes sense.

Test Code

Code should be tested in the same folder that it's written in. See the upstream testing documentation for more information.

Integration tests or other things should be done in an internal subpackage called "integration" or similar.f

Questions and Answers

Why not use pkg/ for packages you intend others to use?

The name pkg is already well-known in the Go ecosystem. It is the folder that compiled packages (not command binaries) go. Using it creates the potential for confusion between code that others are encouraged to use and the meaning that the Go compiler toolchain has.

If a package prefix for publicly available code is really needed, choose a name not already known to the Go compiler toolchain such as "public".

How does this differ from https://github.com/golang-standards/project-layout?

This differs in a few key ways:

• Discourages the use of pkg, because it's obvious if something is publicly available or not if it can be imported outside of the package
• Leaves the development team a lot more agency to decide how to name things

The core philosophy of this layout is that the developers should be able to decide how to put files into the repository.

But I really think I need pkg!

Set up another git repo for those libraries then. If they are so important that other people need to use them, they should probably be in a libraries repo or individual git repos.

Besides, nothing is stopping you from actually using pkg if you want to. Some more experienced go programmers will protest though.

Examples of This in Action

Here are a few examples of views of this layout in action:

• https://github.com/golang/go/tree/master/src
• https://github.com/golang/tools
• https://github.com/PonyvilleFM/aura
• https://github.com/Xe/ln
• https://github.com/goproxyio/goproxy
• https://github.com/heroku/x