Code generation overview

This overview provides information about generating code from your Protobuf files using the Buf CLI, ranging from local generation to generated SDKs hosted in the Buf Schema Registry (BSR).

One of the challenges with Protobuf code generation is the complexity of working with protoc and plugins. Managing and maintaining a stable environment locally on a single machine is hard enough given the complex web of different compiler and plugin versions. The problem is compounded as you scale out code generation across many developers, and often results in a series of ugly bash scripts shared between team members.

Buf's code generation streamlines all of that and is designed to work with whatever your Protobuf setup might be. It has the following advantages over protoc:

• Speed: Compiles 2x faster than protoc.
• Clarity: Allows you to configure your build once and easily edit or reference as needed from source control, instead of memorizing strings of flags.
• Flexibility: Accepts more types of input (protoc only accepts raw .proto files) and the input doesn't have to be local.

In addition, Buf offers several features that ease many of the frustrations of generating client code from Protobuf files, especially in larger organizations:

• Remote plugins: Avoid installing and maintaining protoc plugins locally. Specify our hosted versions instead, right down to the version number, making it easy to standardize across teams or use different versions for different projects, as needed.
• Managed mode: Produce universal APIs by removing user and language specific Protobuf options from .proto files. API consumers can then enable smart defaults for these options in the config file with two lines of code and never think about them again.
• Generated SDKs: Skip the entire code generation process and use generated SDKs. The BSR packages up your modules so they can be consumed via NPM, Maven, and other common dependency management tools. They're created automatically when modules are pushed to the BSR.

The sections below discuss key concepts, local generation (with and without remote plugins), managed mode, and specific invocations of the buf generate command.

Key concepts

protoc

protoc is the compiler included with Protocol Buffers—it invokes a set of plugins to generate code or perform other custom tasks.

input

To the Buf CLI, an input is the representation of the Protobuf files you want to manage. It can take many forms, such as a local directory of .proto files, a module hosted in the Buf Schema Registry, or a .zip file. See the Inputs reference for a full list of inputs that the Buf CLI accepts.

plugins

Plugins are external programs that implement a specific Protobuf interface. They're used by protoc and Buf to generate code or perform other custom processing on Protobuf files. With the Buf CLI, you can either install and use plugins locally, or save time and improve consistency by referencing the remote plugins that we host in the Buf Schema Registry (BSR). They are the same as the ones you locally install.

buf.gen.yaml

This is the Buf config file specifically related to code generation. It's generally placed at the root of your workspace and specifies the plugin to be used for each output, the output path, a set of options that enable you to customize the generated code further.

Generating with local plugins

Generating client code with locally-installed plugins works basically the same as protoc, but faster. In the diagram below, the plugins.local key specifies a local version of protoc-gen-go.

For plugins that are built into protoc, use the plugins.protoc_builtin key and specify the language as the value:

version: v2
plugins:
  - protoc_builtin: cpp
    out: gen/cpp

The protoc_builtin key accepts these values: cpp, csharp, java, objectivec, php, python, or ruby.

Otherwise, use the plugins.local key and provide the name of a plugin on your PATH:

version: v2
plugins:
  - local: protoc-gen-go
    out: gen/go

Generating with remote plugins

As noted above, managing a set of local plugins can complicate your workflow, especially if you're trying to be consistent across teams. A common solution is to rely on custom tooling, which frequently leads to "works on my machine" reproducibility problems. With Buf, you can standardize code generation by specifying the mirrors of the protoc plugins that we host in the BSR. Just point the plugins.remote key to the BSR, and it generates the code.

version: v2
plugins:
  - remote: buf.build/protocolbuffers/go
    out: gen/go

For more information, see Using remote plugins.

Using managed mode

Another issue with code generation is the need define output options that aren't part of the schema. These include fields like package prefixes, namespace information, and so on that aren't part of your API definition but need to be passed to the generator to render the client code correctly.

Normally, API producers hard-code these options in their Protobuf files, but they are really API consumer concerns. Using managed mode allows consumers to easily use smart defaults for these options or override them in the buf.gen.yaml file, freeing the producers to provide a universal API. Using managed mode across an entire project or organization gives you a clear separation of concerns while still providing necessary flexibility to consumers.

The example below demonstrates using managed mode to generate Java code using remote plugins.

syntax = "proto3";

package acme.weather.v1;

option java_multiple_files = true;
option java_outer_classname = "WeatherProto";
option java_package = “com.acme.weather.v1”;

version: v2
plugins:
  - remote: buf.build/protocolbuffers/java
    out: gen/proto/java

syntax = "proto3";

package acme.weather.v1;

version: v2
managed:
  enabled: true
plugins:
  - remote: buf.build/protocolbuffers/java
    out: gen/proto/java

Command-line input examples

The examples below show ways of invoking buf generate for different inputs or subsets of inputs. Generally, you should specify all inputs in the buf.gen.yaml file, but they can also be passed to buf generate at the command line as in the examples below. If you specify an input in buf.gen.yaml and in the command, the command takes precedence and only that input is generated.

See the buf.gen.yaml configuration file documentation to learn how to specify inputs there.

$ buf generate

$ buf generate buf.build/acme/petapis

$ buf generate https://github.com/foo/bar.git

Limit to specific files

By default, the Buf CLI builds all files in the workspace. You can instead manually specify the file or directory paths to build. This is an advanced feature intended to be used for editor or Bazel integration—it's better to let the Buf CLI discover all files under management and handle this for you.

If you only want to generate code for a subset of your input, you can do so via the --path flag:

Only generate for the files in the directories proto/foo and proto/bar

$ buf generate --path proto/foo --path proto/bar

Only generate for the files proto/foo/foo.proto and proto/foo/bar.proto

$ buf generate --path proto/foo/foo.proto --path proto/foo/bar.proto

Only generate for the files in the directory proto/foo in your GitHub repository

$ buf generate https://github.com/foo/bar.git --template data/generate.yaml --path proto/foo

Limit to specific types

By default, buf generates code for all types in your source file(s). However, you can request that it only generate the types that you absolutely need by specifying a list of types you want to include in the buf.gen.yaml, omitting everything else from the generated output. The resulting schema contains these elements and all of their dependencies, according to the following principles. This is especially useful for mobile and web apps, where shipping unused code is expensive.

• All names must be fully-qualified. If any name is unknown, the request will fail and no output will be returned.
• When a message is included, any type references therein, such as from fields that refer to other message or enum types, are also included. If any included message is extendable (a proto2-only feature), its known extensions will be included.
• If a particular method is included, its request and response message types will also be included, but other methods that belong to the same service will not (unless they are also included).
• If a service is included, all of its methods (and their request and response types) are included.
• If any included types have custom options, the definitions for those custom options will also be included.

If you only want to generate code for a subset of your types, you can specify it like this:

version: v1
types:
  include:
    - buf.alpha.lint.v1.IDPaths
plugins:
  - local: protoc-gen-go
    out: gen/go
    opt: paths=source_relative
  - local: go-grpc
    out: gen/go
    opt:
      - paths=source_relative
      - require_unimplemented_servers=false

You can also use the --type flag, which overrides whatever might be configured in buf.gen.yaml.

Only generate the type with a fully qualified name of buf.alpha.lint.v1.IDPaths

$ buf generate --type buf.alpha.lint.v1.IDPaths

Generate to a specific directory

You can generate to a specific output directory using the --output or -o flag. This command generates to the bar/ directory while prepending bar/ to the out directives in the buf.gen.yaml file:

$ buf generate https://github.com/foo/bar.git -o bar

The paths in the config file and the -o flag are interpreted as relative to your current directory, so you can place your config files anywhere.

Generate with a configuration file in another directory

The config file doesn't have to be named buf.gen.yaml, nor does it have to be in the directory where you're invoking buf generate. To specify a config file elsewhere, use the --template flag:

$ buf generate buf.build/acme/petapis --template templates/generation.yaml

Generate using multiple configuration files

For cases where you need to use different output configurations for the same input, we recommend using multiple configuration files with different names.

To use one file for Go and a different file for Java, for example, you could create a buf.gen.go.yaml file and a buf.gen.java.yaml file and use separate commands to generate code:

$ buf generate --template buf.gen.go.yaml
$ buf generate --template buf.gen.java.yaml

Generate without a configuration file

Although we recommend using a configuration file, without one buf generate will assume the input is the current directory. You can also specify the config file as JSON in the invocation. This can be useful when configuring CI/CD builds for an organization when you want to centralize the configuration for all teams.

$ buf generate --template '{"version":"v2","plugins":[{"protoc-builtin":"go","out":"gen/go"}]}'

Related docs

• Try out code generation with the buf generate tutorial
• Learn more about consuming generated SDKs instead of generating code
• Browse the buf generate command reference
• Find out how to create custom plugins for your organization