Original post

Joe Tsai, Damien Neil, and Herbie Ong
2 March 2020

Introduction

We are pleased to announce the release of a major revision of the API for protocol buffers, Google’s language-neutral data interchange format.

Motivations for a new API

The first protocol buffer bindings for Go were announced by Rob Pike in March of 2010. Go 1 would not be released for another two years.

In the decade since that first release, the package has grown and developed along with Go. Its users’ requirements have grown too.

Many people want to write programs that use reflection to examine protocol buffer messages. The reflect package provides a view of Go types and values, but omits information from the protocol buffer type system. For example, we might want to write a function that traverses a log entry and clears any field annotated as containing sensitive data. The annotations are not part of the Go type system.

Another common desire is to use data structures other than the ones generated by the protocol buffer compiler, such as a dynamic message type capable of representing messages whose type is not known at compile time.

We also observed that a frequent source of problems was that the proto.Message interface, which identifies values of generated message types, does very little to describe the behavior of those types. When users create types that implement that interface (often inadvertently by embedding a message in another struct) and pass values of those types to functions expecting a generated message value, programs crash or behave unpredictably.

All three of these problems have a common cause, and a common solution: The Message interface should fully specify the behavior of a message, and functions operating on Message values should freely accept any type that correctly implements the interface.

Since it is not possible to change the existing definition of the Message type while keeping the package API compatible, we decided that it was time to begin work on a new, incompatible major version of the protobuf module.

Today, we’re pleased to release that new module. We hope you like it.

Reflection

Reflection is the flagship feature of the new implementation. Similar to how the reflect package provides a view of Go types and values, the google.golang.org/protobuf/reflect/protoreflect package provides a view of values according to the protocol buffer type system.

A complete description of the protoreflect package would run too long for this post, but let’s look at how we might write the log-scrubbing function we mentioned previously.

First, we’ll write a .proto file defining an extension of the google.protobuf.FieldOptions type so we can annotate fields as containing sensitive information or not.

syntax = "proto3";
import "google/protobuf/descriptor.proto";
package .example.policy;
extend google.protobuf.FieldOptions {
    bool non_sensitive = 50000;
}

We can use this option to mark certain fields as non-sensitive.

message MyMessage {
    string public_name = 1 [(golang.example.policy.non_sensitive) = true];
}

Next, we will write a Go function which accepts an arbitrary message value and removes all the sensitive fields.

// Redact clears every sensitive field in pb.
func Redact(pb proto.Message) {
   // ...
}

This function accepts a proto.Message, an interface type implemented by all generated message types. This type is an alias for one defined in the protoreflect package:

type ProtoMessage interface{
    ProtoReflect() Message
}

To avoid filling up the namespace of generated messages, the interface contains only a single method returning a protoreflect.Message, which provides access to the message contents.

(Why an alias? Because protoreflect.Message has a corresponding method returning the original proto.Message, and we need to avoid an import cycle between the two packages.)

The protoreflect.Message.Range method calls a function for every populated field in a message.

m := pb.ProtoReflect()
m.Range(func(fd protoreflect.FieldDescriptor, v protoreflect.Value) bool {
    // ...
    return true
})

The range function is called with a protoreflect.FieldDescriptor describing the protocol buffer type of the field, and a protoreflect.Value containing the field value.

The protoreflect.FieldDescriptor.Options method returns the field options as a google.protobuf.FieldOptions message.

opts := fd.Options().(*descriptorpb.FieldOptions)

(Why the type assertion? Since the generated descriptorpb package depends on protoreflect, the protoreflect package can’t return the concrete options type without causing an import cycle.)

We can then check the options to see the value of our extension boolean:

if proto.GetExtension(opts, policypb.E_NonSensitive).(bool) {
    return true // don't redact non-sensitive fields
}

Note that we are looking at the field descriptor here, not the field value. The information we’re interested in lies in the protocol buffer type system, not the Go one.

This is also an example of an area where we have simplified the proto package API. The original proto.GetExtension returned both a value and an error. The new proto.GetExtension returns just a value, returning the default value for the field if it is not present. Extension decoding errors are reported at Unmarshal time.

Once we have identified a field that needs redaction, clearing it is simple:

m.Clear(fd)

Putting all the above together, our complete redaction function is:

// Redact clears every sensitive field in pb.
func Redact(pb proto.Message) {
    m := pb.ProtoReflect()
    m.Range(func(fd protoreflect.FieldDescriptor, v protoreflect.Value) bool {
        opts := fd.Options().(*descriptorpb.FieldOptions)
        if proto.GetExtension(opts, policypb.E_NonSensitive).(bool) {
            return true
        }
        m.Clear(fd)
        return true
    })
}

A more complete implementation might recursively descend into message-valued fields. We hope that this simple example gives a taste of protocol buffer reflection and its uses.

Versions

We call the original version of Go protocol buffers APIv1, and the new one APIv2. Because APIv2 is not backwards compatible with APIv1, we need to use different module paths for each.

(These API versions are not the same as the versions of the protocol buffer language: proto1, proto2, and proto3. APIv1 and APIv2 are concrete implementations in Go that both support the proto2 and proto3 language versions.)

The github.com/golang/protobuf module is APIv1.

The google.golang.org/protobuf module is APIv2. We have taken advantage of the need to change the import path to switch to one that is not tied to a specific hosting provider. (We considered google.golang.org/protobuf/v2, to make it clear that this is the second major version of the API, but settled on the shorter path as being the better choice in the long term.)

We know that not all users will move to a new major version of a package at the same rate. Some will switch quickly; others may remain on the old version indefinitely. Even within a single program, some parts may use one API while others use another. It is essential, therefore, that we continue to support programs that use APIv1.

  • github.com/golang/protobuf@v1.3.4 is the most recent pre-APIv2 version of APIv1.
  • github.com/golang/protobuf@v1.4.0 is a version of APIv1 implemented in terms of APIv2. The API is the same, but the underlying implementation is backed by the new one. This version contains functions to convert between the APIv1 and APIv2 proto.Message interfaces to ease the transition between the two.
  • google.golang.org/protobuf@v1.20.0 is APIv2. This module depends upon github.com/golang/protobuf@v1.4.0, so any program which uses APIv2 will automatically pick a version of APIv1 which integrates with it.

(Why start at version v1.20.0? To provide clarity. We do not anticipate APIv1 to ever reach v1.20.0, so the version number alone should be enough to unambiguously differentiate between APIv1 and APIv2.)

We intend to maintain support for APIv1 indefinitely.

This organization ensures that any given program will use only a single protocol buffer implementation, regardless of which API version it uses. It permits programs to adopt the new API gradually, or not at all, while still gaining the advantages of the new implementation. The principle of minimum version selection means that programs may remain on the old implementation until the maintainers choose to update to the new one (either directly, or by updating a dependency).

Additional features of note

The google.golang.org/protobuf/encoding/protojson package converts protocol buffer messages to and from JSON using the canonical JSON mapping, and fixes a number of issues with the old jsonpb package that were difficult to change without causing problems for existing users.

The google.golang.org/protobuf/types/dynamicpb package provides an implementation of proto.Message for messages whose protocol buffer type is derived at runtime.

The google.golang.org/protobuf/testing/protocmp package provides functions to compare protocol buffer messages with the github.com/google/cmp package.

The google.golang.org/protobuf/compiler/protogen package provides support for writing protocol compiler plugins.

Conclusion

The google.golang.org/protobuf module is a major overhaul of Go’s support for protocol buffers, providing first-class support for reflection, custom message implementations, and a cleaned up API surface. We intend to maintain the previous API indefinitely as a wrapper of the new one, allowing users to adopt the new API incrementally at their own pace.

Our goal in this update is to improve upon the benefits of the old API while addressing its shortcomings. As we completed each component of the new implementation, we put it into use within Google’s codebase. This incremental rollout has given us confidence in both the usability of the new API and the performance and correctness of the new implementation. We believe it is production ready.

We are excited about this release and hope that it will serve the Go ecosystem for the next ten years and beyond!