If your system happens to have a microservices architecture, you may find it repetitive to scaffold new services or to add new endpoints. In such scenarios, Protocol Buffer emerges to be an excellent choice for writing API contracts and generating code. However, you sometimes may want to add a custom code but you're not sure how to to that. Actually, it's quite simple with protoc though. In this post, we will walk through how to create a code generator with protoc.
Getting started
Before getting started, make sure you have Go environment ready. It's also important to be familiar with Google protobuf. I highly recommend running through the Go tutorial to generate Go code for any given protocol definition if you're new to the stuff.
While running through the tutorial, you may notice that we need install protoc-gen-go in order to generate Go code. Yep, protoc-gen-go is a plugin of the protoc command written by Google and you can check out its source code. In this guide, we will write a similar program called protoc-gen-my-plugin to generate custom code. As the command to execute protoc-gen-go is like below:
protoc --proto_path=. --go_out=. --go_opt=paths=source_relative foo.proto
The command to execute our plugin will be like:
protoc --proto_path=. --my-plugin_out=. --my-plugin_opt=paths=source_relative foo.proto
In the above command, my-plugin_out specifies the output directory of the generated files and it also tells protoc to use protoc-gen-my-plugin to generate the custom code. my-plugin_opt specifies the option for running the plugin.
Okay, let's write a simple program to test it. At first, I simply use these commands to set up a new Go project:
mkdir protoc-gen-my-plugin
cd protoc-gen-my-plugin
go mod init github.com/bongnv/protoc-gen-my-plugin
export PATH=$PATH:"$(pwd)" # so protoc can find our plugin
Next, create a simple main.go to just print a log:
package main
import (
"log"
)
func main() {
log.Println("protoc-gen-my-plugin is called")
}
We also need to draft a simple foo.proto as an example:
syntax = "proto3";
message Foo {}
After all, we can try our plugin and the log should be printed like below:
% go build && protoc --proto_path=. --my-plugin_out=. --my-plugin_opt=paths=source_relative foo.proto
2021/05/08 16:28:31 protoc-gen-my-plugin is called
Write logic to generate code
As protoc can find our plugin and execute it, now we need to write logic to generate code. It could be a New method for each message to create a new instance of that message to be used as a factory function. The generated code would look simple like below:
func (Foo) New() *Foo {
return &Foo{}
}
By implementation, protoc communicates with our plugin via stdin and stdout. However, protogen.Options already abstracts that logic so we can just use it to simplify the code by providing a callback function:
func main() {
log.Println("protoc-gen-my-plugin is called")
protogen.Options{}.Run(func(plugin *protogen.Plugin) error {
for _, file := range plugin.Files {
if !file.Generate {
continue
}
if err := generateFile(plugin, file); err != nil {
return err
}
}
return nil
})
}
The callback function will take *protogen.Plugin as an input that should contain all necessary information to generate code. In the above example, I skip all proto files that don't require code generation and I call generateFile to generate a source file given a parsed proto file.
Let's look at the function generateFile:
func generateFile(p *protogen.Plugin, f *protogen.File) error {
if len(f.Messages) == 0 {
return nil
}
filename := f.GeneratedFilenamePrefix + "_my_plugin.pb.go"
g := p.NewGeneratedFile(filename, f.GoImportPath)
g.P("// Code generated by protoc-gen-my-plugin. DO NOT EDIT.")
g.P()
g.P("package ", f.GoPackageName)
g.P()
for _, m := range f.Messages {
msgName := m.GoIdent.GoName
g.P("// New creates a new instance of ", msgName, ".")
g.P("func (", msgName, ") New() *", msgName, " {")
g.P("return &", msgName, "{}")
g.P("}")
}
return nil
}
I use f.Messages to retrieve the list of parsed messages from the proto file. We can also use f.Services for the list of parsed services if we need to scaffold a service. NewGeneratedFile here is used to create a new generated file and g.P is used to add a new line into the file. Check their API, there are lots of useful stuff.
Summary
Writing a protoc plugin isn't as complex as it sounds. With the help of protogen.Options and protogen.Plugin we can easily access the parsed information of proto files. And from that, we should be able to generate codes to improve the productivity of the development.
From my experience, it could be handy to scaffold a service including both gRPC and HTTP. It could be used to generate validator code, data entity or DAO to interact with the storage layer.
References: