View Source Protox
protox is an Elixir library to work with Google's Protocol Buffers, versions 2 and 3. It supports both binary and JSON encoding and decoding (well-known types included, except the Any type for the time being).
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.
It's also easy to use: 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.
Given the following protobuf definition, protox will generate a Msg struct:
message Msg{
int32 a = 1;
map<int32, string> b = 2;
}You can then interact with Msg like any Elixir structure:
iex> msg = %Msg{a: 42, b: %{1 => "a map entry"}}
iex> {:ok, iodata} = Msg.encode(msg)
iex> {:ok, iodata} = Msg.json_encode(msg)
iex> binary = # read binary from a socket, a file, etc.
iex> {:ok, msg} = Msg.decode(binary)
iex> json = # read json from a socket, file, etc.
iex> {:ok, msg} = Msg.json_decode(json)You can find here a more involved example with most types.
Table of contents
- Prerequisites
- Installation
- Usage with a textual description
- Usage with files
- Protobuf binary format
- Protobuf JSON format
- Packages and namespaces
- Specify import path
- Unknown fields
- Unsupported features
- Implementation choices
- Generated code reference
- Files generation
- Conformance
- Types mapping
- Benchmarks
- Credits
Prerequisites
- Elixir >= 1.9
- protoc >= 3.0 This dependency is only required at compile-time
protoxuses Google'sprotoc(>= 3.0) to parse.protofiles. It must be available in$PATH. đ You can download it here or you can install it with your favorite package manager (brew install protobuf,apt install protobuf-compiler, etc.). âšī¸ If you choose to generate files,protocwon't be needed to compile these files.
Installation
Add :protox to your list of dependencies in mix.exs:
def deps do
[{:protox, "~> 1.6"}]
endIf you plan to use the JSON encoding, you'll need to add Jason to your dependencies:
def deps do
[
{:protox, "~> 1.6"},
{:jason, "~> 1.2"}
]
endUsage with an inlined textual description
The following example generates two modules: Baz and Foo from a textual description:
defmodule MyModule do
use Protox, schema: """
syntax = "proto3";
message Baz {
}
message Foo {
int32 a = 1;
map<int32, Baz> b = 2;
}
"""
endâšī¸ 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 MyModule do
use Protox, files: [
"./defs/foo.proto",
"./defs/bar.proto",
"./defs/baz/fiz.proto",
]
endProtobuf binary format
Encode
Here's how to create and encode a new message to binary protobuf:
iex> msg = %Fiz.Foo{a: 3, b: %{1 => %Fiz.Baz{}}}
iex> {:ok, iodata} = Protox.encode(msg)Or, with throwing style:
iex> iodata = Protox.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)
iex> iodata = Fiz.Foo.encode!(msg)âšī¸ Note that encode/1 returns an IO data for efficiency reasons. Such IO data can be used
directly with files or sockets write operations:
iex> {:ok, iodata} = Protox.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)
:okđ You can use :binary.list_to_bin/1 or IO.iodata_to_binary to get a binary:
iex> %Fiz.Foo{a: 3, b: %{1 => %Fiz.Baz{}}} |> Protox.encode!() |> :binary.list_to_bin()
<<8, 3, 18, 4, 8, 1, 18, 0>>Decode
Here's how to decode a message from binary protobuf:
iex> {:ok, msg} = Protox.decode(<<8, 3, 18, 4, 8, 1, 18, 0>>, Fiz.Foo)Or, with throwing style:
iex> msg = Protox.decode!(<<8, 3, 18, 4, 8, 1, 18, 0>>, Fiz.Foo)It's also possible to call decode/1 and decode!/1 directly on the generated structures:
iex> {:ok, msg} = Fiz.Foo.decode(<<8, 3, 18, 4, 8, 1, 18, 0>>)
iex> msg = Fiz.Foo.decode!(<<8, 3, 18, 4, 8, 1, 18, 0>>)Protobuf JSON format
protox implements the Google's JSON specification.
Encode
Here's how to encode a message to JSON, exported as IO data:
iex> msg = %Fiz.Foo{a: 42}
iex> {:ok, iodata} = Protox.json_encode(msg)
{:ok, ["{", ["\"a\"", ":", "42"], "}"]}Or, with throwing style:
iex> msg = %Fiz.Foo{a: 42}
iex> iodata = Protox.json_encode!(msg)
["{", ["\"a\"", ":", "42"], "}"]It's also possible to call json_encode and json_encode! directly on the generated structures:
iex> {:ok, iodata} = Fiz.Foo.json_encode(msg)
iex> iodata = Fiz.Foo.json_encode!(msg)Decode
Here's how to decode JSON to a message:
iex> Protox.json_decode("{\"a\":42}", Fiz.Foo)
{:ok, %Fiz.Foo{__uf__: [], a: 42, b: %{}}}Or, with throwing style:
iex> Protox.json_decode!("{\"a\":42}", Fiz.Foo)
%Fiz.Foo{__uf__: [], a: 42, b: %{}}It's also possible to call json_decode and json_decode! directly on the generated structures:
iex> Fiz.Foo.json_decode("{\"a\":42}")
iex> Fiz.Foo.json_decode!("{\"a\":42}")JSON library configuration
By default, protox uses Jason to encode values to JSON (mostly to escape strings). You can also use Poison:
iex> Protox.json_decode!(iodata, Fiz.Foo, json_library: Protox.Poison)
iex> Protox.json_encode!(msg, json_library: Protox.Poison)âšī¸ You can use any other library by implementing the Protox.JsonLibrary behaviour.
đ Don't forget to add the chosen library to the list of dependencies in mix.exs.
Well-known types
Note that protox does not completely support the Any well-know type: it will be encoded and decoded like a regular message, rather than with the custom encoding specified in the JSON specification.
Packages and namespaces
Packages
Protobuf provides a package directive:
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 with the namespace option to all generated modules:
defmodule Bar do
use Protox, schema: """
syntax = "proto3";
package abc;
message Msg {
int32 a = 1;
}
""",
namespace: MyApp
endIn this example, the module MyApp.Abc.Msg is generated:
iex> msg = %MyApp.Abc.Msg{a: 42}Specify import path
An import path can be specified using the path: or paths: options that respectively specify the directory or directories in which to search for imports:
defmodule Baz do
use Protox,
files: [
"./defs/prefix/foo.proto",
"./defs/prefix/bar/bar.proto",
],
path: "./defs"
endIf multiple search paths are needed:
defmodule Baz do
use Protox,
files: [
"./defs1/prefix/foo.proto",
"./defs1/prefix/bar.proto",
"./defs2/prefix/baz/baz.proto"
],
paths: [
"./defs1",
"./defs2"
]
endIt 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 enables backwards 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 unknown_fields/1 function 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.
Disable support of unknown fields
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
endâšī¸ 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. This also means that unkown fields won't be serialized back.
Unsupported features
- The Any well-known type is partially supported: you can manually unpack the embedded message right after decoding and conversely pack it right before encoding;
- Groups (deprecated in protobuf);
- All options other than
packedanddefaultare 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!(%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,
:BARis always used if the value1is 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 generateddefault/1function 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 nilIt 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
nilor not.For Protobuf 3, unset fields are mapped to their default values. However, if you use the
optionalkeyword (available in protoc version 3.15 and higher), then unset fields will be mapped tonil:defmodule Bar do use Protox, schema: """ syntax = "proto3"; message Foo { int32 a = 1; optional int32 b = 2; } """ end iex> Foo.default(:a) {:ok, 0} iex> %Foo{}.a 0 iex> Foo.default(:b) {:error, :no_default_value} iex> %Foo{}.b nil
Messages and enums names: they are converted using the
Macro.camelize/1function. Thus, in the following example,non_camel_messagebecomesNonCamelMessage, but the fieldnon_camel_fieldis 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 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/samples test/samples/messages.proto test/samples/proto2.proto
The generated file will be usable in any project as long as protox is declared in the dependencies as it needs functions from the protox runtime.
Options
--output-pathThe path to the file to be generated or to the destination folder when generating multiple files.--include-pathSpecifies the import path. If multiple include paths are needed, add more--include-pathoptions.--multiple-filesGenerates one file per module. In this case,--output-pathmust point to a directory. It's useful for definitions with a lot of messages as Elixir will be able to parallelize the compilation of the generated modules.--namespacePrepends a namespace to all generated modules.--keep-unknown-fields=[true|false]Toggles support of unknown fields. Default totrue.--generate-defs-funs=[true|false]Generates deprecated functionsdefs/0anddefs_by_name/0. Default totruefor backward compatibility, but it's safe to set it tofalseif you don't use these functions (and it reduces the size of the generated code).
Conformance
The protox library has been thoroughly tested using the conformance checker provided by Google.
Here's how to launch the conformance tests:
Get conformance-test-runner sources.
Compile conformance-test-runner (macOS and Linux only):
tar xf protobuf-3.18.0.tar.gz && cd protobuf-3.18.0 && ./autogen.sh && ./configure && make -j && cd conformance && make -jLaunch the conformance tests:
mix protox.conformance --runner=/path/to/protobuf-3.18.0/conformance/conformance-test-runnerA report will be generated in the directory
conformance_reportand the following text should be displayed:CONFORMANCE TEST BEGIN ==================================== CONFORMANCE SUITE PASSED: 1996 successes, 0 skipped, 21 expected failures, 0 unexpected failures. CONFORMANCE TEST BEGIN ==================================== CONFORMANCE SUITE PASSED: 0 successes, 120 skipped, 0 expected failures, 0 unexpected failures.You can alternatively launch these conformance tests with
mix testby setting thePROTOBUF_CONFORMANCE_RUNNERenvironment variable and including theconformancetag:PROTOBUF_CONFORMANCE_RUNNER=/path/to/conformance-test-runner MIX_ENV=test mix test --include conformance
Skipped conformance tests
You may have noticed that there are 21 expected failures. Indeed, we removed on purpose some conformance tests that protox can't currently pass. Here are the reasons why:
- Any is not yet supported by
protox; - We could not find the specification for the protobuf2 case of field name extensions when decoding from JSON.
The exact list of skipped tests is here.
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' | :nan |
double | float() | :infinity | :'-infinity' | :nan |
bool | 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.exsCredits
Both gpb and exprotobuf were very useful in understanding how to implement Protocol Buffers.