Protox

Elixir CI

Protox is an Elixir library to work with Google’s Protocol Buffers (aka protobuf), versions 2 and 3.

The primary objective of protox is reliability: it uses property based testing and has a near 100% code coverage. Also, using mutation testing with the invaluable help of Muzak pro, the quality of the protox test suite has been validated. Therefore, protox passes all the tests of the conformance checker provided by Google. See Conformance section for more information.

This library is easy to use: you just point to the *.proto files or give the schema to the Protox macro, no need to generate any file! However, should you need to generate files, a mix task is available (see Files generation).

Protox provides a full-blown Elixir experience with protobuf messages. For instance, given the following protobuf msg.proto file:

syntax = "proto3";

message Msg{
  int32 a = 1;
  map<int32, string> b = 2;
}

You can interact with Msg as if it were a native Elixir structure. For example, note how the map b is translated into a map:

iex> msg = %Msg{a: 42, b: %{1 => "a map entry"}}
iex> {:ok, iodata} = Protox.Encode.encode(msg) # or Msg.encode(msg)
...
iex> binary = # read binary from a socket, a file, etc.
iex> {:ok, msg} = Msg.decode(binary)

You can find here a more involved example with most types.

Protox

Prerequisites

Protox uses Google’s protoc (>= 3.0) to parse .proto files. It must be available in $PATH. You can download it here or you can install it with your favorite package manager (brew install protobuf, apt install protobuf-compiler, etc.).

This dependency is only required at compile-time.

Installation

Add :protox to your list of dependencies in mix.exs:

def deps do
  [{:protox, "~> 1.3"}]
end

Usage with a textual description

Here’s how to generate the modules from a textual description:

defmodule Bar do
  use Protox, schema: """
  syntax = "proto3";

  package fiz;

  message Baz {
  }

  message Foo {
    int32 a = 1;
    map<int32, Baz> b = 2;
  }
  """
end

This example will generate two modules: Fiz.Baz and Fiz.Foo. Note that the module in which the Protox macro is called is completely ignored and therefore does not appear in the names of the generated modules.

Usage with files

Here’s how to generate the modules from a set of files:

defmodule Foo do
  use Protox, files: [
    "./defs/foo.proto",
    "./defs/bar.proto",
    "./defs/baz/fiz.proto",
  ]
end

Again, the module in which the Protox macro is called is completely ignored.

Encode

Here’s how to create and encode a new message:

iex> msg = %Fiz.Foo{a: 3, b: %{1 => %Fiz.Baz{}}}
iex> {:ok, iodata} = Protox.Encode.encode(msg)

Or, with throwing style:

iex> iodata = Protox.Encode.encode!(msg)

It’s also possible to call encode/1 and encode!/1 directly on the generated structures:

iex> {:ok, iodata} = Fiz.Foo.encode(msg)

Note that encode/1 returns an IO data, not a binary, for efficiency reasons. Such IO data can be used directly with files or sockets write operations, and therefore you don’t need to transform them:

iex> {:ok, iodata} = Protox.Encode.encode(%Fiz.Foo{a: 3, b: %{1 => %Fiz.Baz{}}})
[[[], <<18>>, <<4>>, "\b", <<1>>, <<18>>, <<0>>], "\b", <<3>>]

iex> {:ok, file} = File.open("msg.bin", [:write])
{:ok, #PID<0.1023.0>}

iex> IO.binwrite(file, iodata)
:ok

However, you can use :binary.list_to_bin/1 or IO.iodata_to_binary to get a binary should the need arises:

iex> %Fiz.Foo{a: 3, b: %{1 => %Fiz.Baz{}}} |> Protox.Encode.encode!() |> :binary.list_to_bin()
<<8, 3, 18, 4, 8, 1, 18, 0>>

Decode

Here’s how to decode a message from a binary:

iex> {:ok, msg} = Fiz.Foo.decode(<<8, 3, 18, 4, 8, 1, 18, 0>>)

Or, with throwing style:

iex> msg = Fiz.Foo.decode!(<<8, 3, 18, 4, 8, 1, 18, 0>>)

Packages and namespaces

Packages

Protobuf provides a packagedirective:

syntax = "proto3";

package abc.def;

message Baz {
}

Modules generated by protox will include this package declaration. Thus, the example above will be translated to Abc.Def.Baz (note the camelization of package abc.def to Abc.Def).

Prepend namespaces

In addition, protox provides the possibility to prepend a namespace to all generated modules:

defmodule Bar do
  use Protox, schema: """
    syntax = "proto3";

    package abc;

    message Msg {
        int32 a = 1;
      }
    """,
    namespace: MyApp
end

In this example, the module MyApp.Abc.Msg is generated:

iex> msg = %MyApp.Msg{a: 42}

It’s useful to make the generated code appear as being part of your code structure.

Specify import path

An import path can be specified using the path: option that specifies the directory in which to search for import:

defmodule Baz do
  use Protox,
    files: [
      "./defs/prefix/foo.proto",
      "./defs/prefix/bar/bar.proto",
    ],
    path: "./defs"
end

It corresponds to the -I option of protoc.

Unknown fields

Unknown fields are fields that are present on the wire but which do not correspond to an entry in the protobuf definition. Typically, it occurs when the sender has a newer version of the protobuf definition. It makes possible to have backward compatibility as the receiver with an old version of the protobuf definition will still be able to decode old fields.

When unknown fields are encountered at decoding time, they are kept in the decoded message. It’s possible to access them with the function unknown_fields/1 defined with the message.

iex> msg = Msg.decode!(<<8, 42, 42, 4, 121, 97, 121, 101, 136, 241, 4, 83>>)
%Msg{a: 42, b: "", z: -42, __uf__: [{5, 2, <<121, 97, 121, 101>>}]}

iex> Msg.unknown_fields(msg)
[{5, 2, <<121, 97, 121, 101>>}]

You must always use unknown_fields/1 as the name of the field (e.g. __uf__ in the above example) is generated at compile-time to avoid collision with the actual fields of the Protobuf message. This function returns a list of tuples {tag, wire_type, bytes}. For more information, please see protobuf encoding guide.

When you encode a message that contains unknown fields, they will be reencoded in the serialized output.

Finally, you can deactivate the support of unknown fields by setting the :keep_unknown_fields option to false:

defmodule Baz do
  use Protox,
    schema: """
    syntax = "proto3";

    message Sub {
      int32 a = 1;
      string b = 2;
    }
    """,
    keep_unknown_fields: false
end

Note that protox will still correctly parse unknown fields, they just won’t be added to the structure and you won’t be able to access them.

Unsupported features

Protobuf 3 JSON mapping
Groups (deprecated in protobuf)
All options other than packed and default are ignored as they concern other languages implementation details.

Implementation choices

This library enforces the presence of required fields (Protobuf 2). Therefore an error is raised when encoding or decoding a message with a missing required field:

  defmodule Bar do
    use Protox, schema: """
      syntax = "proto2";

      message Required {
        required int32 a = 1;
      }
    """
  end

  iex> Protox.Encode.encode!(%Required{})
  ** (Protox.RequiredFieldsError) Some required fields are not set: [:a]

  iex> Required.decode!(<<>>)
  ** (Protox.RequiredFieldsError) Some required fields are not set: [:a]

When decoding enum aliases, the last encountered constant is used. For instance, in the following example, :BAR is always used if the value 1 is read on the wire:
```
  enum E {
    option allow_alias = true;
    FOO = 0;
    BAZ = 1;
    BAR = 1;
  }
```

Unset optionals

For Protobuf 2, unset optional fields are mapped to nil. You can use the generated default/1 function to get the default value of a field:

  defmodule Bar do
    use Protox,
    schema: """
      syntax = "proto2";

      message Foo {
        optional int32 a = 1 [default = 42];
      }
    """
  end

  iex> Foo.default(:a)
  {:ok, 42}

  iex> %Foo{}.a
  nil

It means that if you need to know if a field has been set by the sender, you just have to test if its value is nil or not.

For Protobuf 3, unset optional fields are mapped to their default values, as mandated by the Protobuf spec:

  defmodule Bar do
    use Protox,
    schema: """
      syntax = "proto3";

      message Foo {
        int32 a = 1;
      }
    """
  end

  iex> Foo.default(:a)
  {:ok, 0}

  iex> %Foo{}.a
  0

Messages and enums names: names are converted using the Macro.camelize/1 function. Thus, in the following example, non_camel_message becomes NonCamelMessage, but the field non_camel_field is left unchanged:

  defmodule Bar do
    use Protox,
    schema: """
      syntax = "proto3";

      message non_camel_message {
      }

      message CamelMessage {
        int32 non_camel_field = 1;
      }
    """
  end


  iex> msg = %NonCamelMessage{}
  %NonCamelMessage{__uf__: []}

  iex> msg = %CamelMessage{}
  %CamelMessage{__uf__: [], non_camel_field: 0}

Generated code reference

The detailed reference of the generated code is available here.

Files generation

It’s also possible to generate a file that will contain all code corresponding to the protobuf messages:

MIX_ENV=prod mix protox.generate --output-path=/path/to/message.ex --include-path=. test/messages.proto test/samples/proto2.proto

The --include-path option is the same as the option described in section Specify import path.

The generated file will be usable in any project as long as protox is declared in the dependencies (the generated file still needs functions from the protox runtime).

If you have large protobuf files, you can use the --multiple-files option to generate one file per module.

mkdir generated
MIX_ENV=prod mix protox.generate --multiple-files --output-path=generated --include-path=. test/messages.proto test/samples/proto2.proto

Doing so, Elixir will be able to parallelize the compilation of generated modules.

It is also possible to prepend a namespace to all generated modules using the --namespace option. More information is available in section Prepend namespaces.

Finally, you can pass the option --keep-unknown-fields=false to remove support of unknown fields. See this section for more information.

Conformance

The protox library has been thoroughly tested using the conformance checker provided by Google. Note that only the binary part is tested as protox supports only this format. For instance, JSON tests are skipped.

Here’s how to launch the conformance test:

Get conformance-test-runner sources.
Compile conformance-test-runner (macOS and Linux only): tar xf v3.14.0.tar.gz && cd protobuf-3.14.0 && ./autogen.sh && ./configure && make -j && cd conformance && make -j.

Run mix protox.conformance --runner=/path/to/protobuf-3.14.0/conformance/conformance-test-runner. A report will be generated in a directory conformance_report. If everything’s fine, the following text should be displayed:

CONFORMANCE TEST BEGIN ====================================

CONFORMANCE SUITE PASSED: 1302 successes, 711 skipped, 0 expected failures, 0 unexpected failures.


CONFORMANCE TEST BEGIN ====================================

CONFORMANCE SUITE PASSED: 0 successes, 119 skipped, 0 expected failures, 0 unexpected failures.

You can alternatively launch these conformance tests with mix test by setting the PROTOBUF_CONFORMANCE_RUNNER environment variable and including the conformance tag:

   PROTOBUF_CONFORMANCE_RUNNER=./protobuf-3.14.0/conformance/conformance-test-runner MIX_ENV=test mix test --include conformance

Types mapping

The following table shows how Protobuf types are mapped to Elixir’s ones.

Protobuf | Elixir ———- | ————- int32 | integer()int64 | integer()uint32 | integer()uint64 | integer()sint32 | integer()sint64 | integer()fixed32 | integer()fixed64 | integer()sfixed32 | integer()sfixed64 | integer()float | float() \| :infinity \| :'-infinity' \| :nandouble | float() \| :infinity \| :'-infinity' \| :nanbool | boolean()string | String.t()bytes | binary()repeated | list(value_type) where value_type is the type of the repeated field map | map()oneof | {atom(), value_type} where atom() is the type of the set field and where value_type is the type of the set field enum | atom() \| integer()message | struct()

Benchmarks

You can launch benchmarks to see how Protox perform:

mix run ./benchmarks/generate_payloads.exs # first time only, generates random payloads
mix run ./benchmarks/run.exs --lib=./benchmarks/protox.exs
mix run ./benchmarks/load.exs

Credits

Both gpb and exprotobuf were very useful in understanding how to implement Protocol Buffers.