#Go
#Deep dive
John McBride
8 mins read
Last week, the OpenSauced engineering team released the Pizza CLI, a powerful and composable command-line tool for generating CODEOWNER files and integrating with the OpenSauced platform. Building robust command-line tools may seem straightforward, but without careful planning and thoughtful paradigms, CLIs can quickly become tangled messes of code that are difficult to maintain and riddled with bugs. In this blog post, we'll take a deep dive into how we built this CLI using Go, how we organize our commands using Cobra, and how our lean engineering team iterates quickly to build powerful functionality.
The Pizza CLI is a Go command-line tool that leverages several standard libraries. Go’s simplicity, speed, and systems programming focus make it an ideal choice for building CLIs. At its core, the Pizza-CLI uses spf13/cobra, a CLI bootstrapping library in Go, to organize and manage the entire tree of commands.
You can think of Cobra as the scaffolding that makes a command-line interface itself work, enables all the flags to function consistently, and handles communicating to users via help messages and automated documentation.
One of the first (and biggest) challenges when building a Cobra-based Go CLI is how to structure all your code and files. Contrary to popular belief, there is no prescribed way to do this in Go. Neither the go build
command nor the gofmt
utility will complain about how you name your packages or organize your directories. This is one of the best parts of Go: its simplicity and power make it easy to define structures that work for you and your engineering team!
Ultimately, in my opinion, it's best to think of and structure a Cobra-based Go codebase as a tree of commands:
├── Root command
│ ├── Child command
│ ├── Child command
│ │ └── Grandchild command
At the base of the tree is the root command: this is the anchor for your entire CLI application and will get the name of your CLI. Attached as child commands, you’ll have a tree of branching logic that informs the structure of how your entire CLI flow works.
One of the things that’s incredibly easy to miss when building CLIs is the user experience. I typically recommend people follow a “root verb noun” paradigm when building commands and child-command structures since it flows logically and leads to excellent user experiences.
For example, in Kubectl, you’ll see this paradigm everywhere: “kubectl get pods”, “kubectl apply …“, or “kubectl label pods …” This ensures a sensical flow to how users will interact with your command line application and helps a lot when talking about commands with other people.
In the end, this structure and suggestion can inform how you organize your files and directories, but again, ultimately it’s up to you to determine how you structure your CLI and present the flow to end-users.
In the Pizza CLI, we have a well defined structure where child commands (and subsequent grandchildren of those child commands) live. Under the cmd
directory in their own packages, each command gets its own implementation. The root command scaffolding exists in a pkg/utils
directory since it's useful to think of the root command as a top level utility used by main.go
, rather than a command that might need a lot of maintenance. Typically, in your root command Go implementation, you’ll have a lot of boilerplate setting things up that you won’t touch much so it’s nice to get that stuff out of the way.
Here's a simplified view of our directory structure:
├── main.go
├── pkg/
│ ├── utils/
│ │ └── root.go
├── cmd/
│ ├── Child command dir
│ ├── Child command dir
│ │ └── Grandchild command dir
This structure allows for clear separation of concerns and makes it easier to maintain and extend the CLI as it grows and as we add more commands.
One of the main libraries we use in the Pizza-CLI is the go-git library, a pure git implementation in Go that is highly extensible. During CODEOWNERS
generation, this library enables us to iterate the git ref log, look at code diffs, and determine which git authors are associated with the configured attributions defined by a user.
Iterating the git ref log of a local git repo is actually pretty simple:
// 1. Open the local git repository
repo, err := git.PlainOpen("/path/to/your/repo")
if err != nil {
panic("could not open git repository”)
}
// 2. Get the HEAD reference for the local git repo
head, err := repo.Head()
if err != nil {
panic("could not get repo head”)
}
// 3. Create a git ref log iterator based on some options
commitIter, err := repo.Log(&git.LogOptions{
From: head.Hash(),
})
if err != nil {
panic("could not get repo log iterator”)
}
defer commitIter.Close()
// 4. Iterate through the commit history
err = commitIter.ForEach(func(commit *object.Commit) error {
// process each commit as the iterator iterates them
return nil
})
if err != nil {
panic("could not process commit iterator”)
}
If you’re building a Git based application, I definitely recommend using go-git: it’s fast, integrates well within the Go ecosystem, and can be used to do all sorts of things!
Our engineering and product team is deeply invested in bringing the best possible command line experience to our end users: this means we’ve taken steps to integrate anonymized telemetry that can report to Posthog on usage and errors out in the wild. This has allowed us to fix the most important bugs first, iterate quickly on popular feature requests, and understand how our users are using the CLI.
Posthog has a first party library in Go that supports this exact functionality. First, we define a Posthog client:
import "github.com/posthog/posthog-go"
// PosthogCliClient is a wrapper around the posthog-go client and is used as a
// API entrypoint for sending OpenSauced telemetry data for CLI commands
type PosthogCliClient struct {
// client is the Posthog Go client
client posthog.Client
// activated denotes if the user has enabled or disabled telemetry
activated bool
// uniqueID is the user's unique, anonymous identifier
uniqueID string
}
Then, after initializing a new client, we can use it through the various struct methods we’ve defined. For example, when logging into the OpenSauced platform, we capture specific information on a successful login:
// CaptureLogin gathers telemetry on users who log into OpenSauced via the CLI
func (p *PosthogCliClient) CaptureLogin(username string) error {
if p.activated {
return p.client.Enqueue(posthog.Capture{
DistinctId: username,
Event: "pizza_cli_user_logged_in",
})
}
return nil
}
During command execution, the various “capture” functions get called to capture error paths, happy paths, etc.
For the anonymized IDs, we use Google’s excellent UUID Go library:
newUUID := uuid.New().String()
These UUIDs get stored locally on end users machines as JSON under their home directory: ~/.pizza-cli/telemtry.json
. This gives the end user complete authority and autonomy to delete this telemetry data if they want (or disable telemetry altogether through configuration options!) to ensure they’re staying anonymous when using the CLI.
Our lean engineering team follows an iterative development process, focusing on delivering small, testable features rapidly. Typically, we do this through GitHub issues, pull requests, milestones, and projects. We use Go's built-in testing framework extensively, writing unit tests for individual functions and integration tests for entire commands.
Unfortunately, Go’s standard testing library doesn’t have great assertion functionality out of the box. It’s easy enough to use “==” or other operands, but most of the time, when going back and reading through tests, it’s nice to be able to eyeball what’s going on with assertions like “assert.Equal” or “assert.Nil”.
We’ve integrated the excellent testify library with its “assert” functionality to allow for smoother test implementation:
config, _, err := LoadConfig(nonExistentPath)
require.Error(t, err)
assert.Nil(t, config)
We heavily use Just at OpenSauced, a command runner utility, much like GNU’s “make”, for easily executing small scripts. This has enabled us to quickly onramp new team members or community members to our Go ecosystem since building and testing is as simple as “just build” or “just test”!
For example, to create a simple build utility in Just, within a justfile, we can have:
build:
go build main.go -o build/pizza
Which will build a Go binary into the build/ directory. Now, building locally is as simple as executing a “just” command.
But we’ve been able to integrate more functionality into using Just and have made it a cornerstone of how our entire build, test, and development framework is executed. For example, to build a binary for the local architecture with injected build time variables (like the sha the binary was built against, the version, the date time, etc.), we can use the local environment and run extra steps in the script before executing the “go build”:
build:
#!/usr/bin/env sh
echo "Building for local arch"
export VERSION="${RELEASE_TAG_VERSION:-dev}"
export DATETIME=$(date -u +"%Y-%m-%d-%H:%M:%S")
export SHA=$(git rev-parse HEAD)
go build \
-ldflags="-s -w \
-X 'github.com/open-sauced/pizza-cli/pkg/utils.Version=${VERSION}' \
-X 'github.com/open-sauced/pizza-cli/pkg/utils.Sha=${SHA}' \
-X 'github.com/open-sauced/pizza-cli/pkg/utils.Datetime=${DATETIME}' \
-X 'github.com/open-sauced/pizza-cli/pkg/utils.writeOnlyPublicPosthogKey=${POSTHOG_PUBLIC_API_KEY}'" \
-o build/pizza
We’ve even extended this to enable cross architecture and OS build: Go uses the GOARCH and GOOS env vars to know which CPU architecture and operating system to build against. To build other variants, we can create specific Just commands for that:
# Builds for Darwin linux (i.e., MacOS) on arm64 architecture (i.e. Apple silicon)
build-darwin-arm64:
#!/usr/bin/env sh
echo "Building darwin arm64"
export VERSION="${RELEASE_TAG_VERSION:-dev}"
export DATETIME=$(date -u +"%Y-%m-%d-%H:%M:%S")
export SHA=$(git rev-parse HEAD)
export CGO_ENABLED=0
export GOOS="darwin"
export GOARCH="arm64"
go build \
-ldflags="-s -w \
-X 'github.com/open-sauced/pizza-cli/pkg/utils.Version=${VERSION}' \
-X 'github.com/open-sauced/pizza-cli/pkg/utils.Sha=${SHA}' \
-X 'github.com/open-sauced/pizza-cli/pkg/utils.Datetime=${DATETIME}' \
-X 'github.com/open-sauced/pizza-cli/pkg/utils.writeOnlyPublicPosthogKey=${POSTHOG_PUBLIC_API_KEY}'" \
-o build/pizza-${GOOS}-${GOARCH}
Building the Pizza CLI using Go and Cobra has been an exciting journey and we’re thrilled to share it with you. The combination of Go's performance and simplicity with Cobra's powerful command structuring has allowed us to create a tool that's not only robust and powerful, but also user-friendly and maintainable.
We invite you to explore the Pizza CLI GitHub repository, try out the tool, and let us know your thoughts. Your feedback and contributions are invaluable as we work to make code ownership management easier for development teams everywhere!
John McBride is a Sr Software Engineer at OpenSauced where he is heading up infrastructure and AI/ML. He's previously worked on Linux based operating systems at AWS, Kubernetes products at VMware, and Cloud Foundry while at Pivotal. He has years of experience building high-scale systems in a number of languages and frameworks. He lives in Denver Colorado with his wife where he enjoys going on runs with his two dogs, Arlo & Zoey.
Recent Posts
BekahHW
6 mins read
Understand the important role of transparency in open source through the PearAI and Continue.dev case study.
#pizza-cli
BekahHW
6 mins read
Discover the hidden costs of siloed knowledge in developer culture and how embracing shared ownership and tools like the CODEOWNERS file can enhance t...