package ocaml-protoc-plugin

Ocaml protoc plugin

The goal of Ocaml protoc plugin is to create an up to date plugin for the google protobuf compiler (protoc) to generate Ocaml types and serialization and de-serialization function from a .proto file.

The main features include:

• Messages are mapped to idiomatic OCaml types, using modules

• Support service descriptions

• proto3 compliant

• proto2 compliant

• Supports includes

• Supports proto2 extensions

• Builtin support for google well known types

• Configurable annotations for all generated types

Comparison with other OCaml protobuf handlers.

[^1] Ocaml-bp has a sister project Ocaml-bp-plugin which emit Ocaml-pb definitions from a .proto. The plugin parses files are proto2 Ocaml type definitions (all fields are option types), and repeated fields are not packed by default.

[^2] Groups has been deprecated by google and should not be used.

Types

Basic types are mapped trivially to Ocaml types:

Primitive types:

[^3] The plugin supports changing the type for scalar types to int/int64/int32. See options section below.

A message declaration is compiled to a module with a record type t. However, messages without any fields are mapped to unit.

Packages are trivially mapped to modules. Included proto files (<name>.proto) are assumed to have been compiled to <name>.ml, and types in included proto files are referenced by their fill name.

Compound types are mapped like:

Proto2 type support

The specification for proto2 states that when deserializing a message, fields which are not transmitted should be set the the default value (either 0, or the value of the default option).

However, It seems to be the norm for proto2, that it should be possible to determine if a field was transmitted or not. Therefore all non-repeated fields in proto2 are default - unless it has a default value, or is a required field.

The specification is vague when it comes to serializing fields for proto2 syntax. (i.e. should fields with default values be serialized?). This implementation chooses to always serialize values, default or otherwise, for all fields which are set (i.e. Some x). This differs from proto3, which explicitly states that fields with default values should not be transmitted).

Invocation

If the plugin is available in the path as protoc-gen-ocaml, then you can generate the Ocaml code by running

  protoc --ocaml_out=. --ocaml_opt=<options> file.proto

Options

Options control the code/types generated.

Parameters are separated by ;

If protoc-gen-ocaml is not located in the path, it is possible to specify the exact path to the plugin:

protoc --plugin=protoc-gen-ocaml=../plugin/ocaml-protocol-plugin.exe --ocaml_out=. <file>.proto

Older versions of protoc

It seems that the --ocaml_opt flag may not be supported by older versions of the proto compiler. As an alternative, options can also be passed with the --ocaml_out flag:

protoc --plugin=protoc-gen-ocaml=../plugin/ocaml.exe --ocaml_out=annot=debug;[@@deriving show { with_path = false }, eq]:. <file>.proto

Mangle generated names

Idiomatic protobuf names are somewhat alien to Ocaml. Ocaml_protoc_plugin has an option to mangle protobuf names into somewhat more Ocaml idiomatic names. When this option is set (see below), names are mangled to snake case as described in the table below:

protoc cannot guarantee that names do not clash when mangling is enabled. If a name clash is detected (eg. SomeMessage and some_message exists in same file) an apostrophe is appended to the name to make sure names are unique.

The algorithm for converting CamelCased names to snake_case is done by injecting an underscore between any lowercase and uppercase character and then lowercasing the result.

Setting mangle option

Name mangling option can only be controlled from within the protobuf specification file. This is needed as protobuf files may reference each other and it its important to make sure that included names are referenced correctly across compilation units (and invocations of protoc).

To set the option use:

// This can be placed in a common file and then included
import "google/protobuf/descriptor.proto";
message options { bool mangle_names = 1; }
extend google.protobuf.FileOptions {
    options ocaml_options = 1074;
}

// This option controls name generation. If true names are converted
// into more ocaml ideomatic names
option (ocaml_options) = { mangle_names:true };

// This message will be mapped to module name My_proto_message
message MyProtoMessage { }

Using dune

Below is a dune rule for generating code for test.proto. The google_include target is used to determine the base include path for google protobuf well known types.

(rule
 (targets google_include)
 (action (with-stdout-to %{targets} (system "[ -d /usr/include/google/protobuf ] && echo /usr/include || echo /usr/local/include" ))))

(rule
 (targets test.ml)
 (deps
  (:proto test.proto))
 (action
  (run protoc -I %{read-lines:google_include} -I .  "--ocaml_opt=annot=[@@deriving show { with_path = false }, eq]" --ocaml_out=. %{proto})))

Service interface

Service interfaces create a module with values that just references the request and reply pair. These binding can then be used with function in Protobuf.Service.

The call function will take a string -> string function, which implement message sending -> receiving.

The service function is a string -> string function which takes a handler working over the actual message types.

Proto2 extensions

Proto2 extensions allows for messages to be extended. For each extending field, the plugin create a module with a get and set function for reading/writing extension fields.

Below is an example on how to set and get extension fields

// file: ext.proto
syntax = "proto2";
message Foo {
  required uint32 i = 1;
  extensions 100 to 200;

}
extend Foo {
  optional uint32 bar = 128;
  optional string baz = 129;
}

(* file: test.ml *)

open Extensions

(* Set extensions *)
let _ =
  let foo = Foo.{ i = 31; extensions' = Ocaml_protoc_plugin.Extensions.default } in
  let foo_with_bar = Bar.set foo (Some 42) in
  let foo_with_baz = Baz.set foo (Some "Test String") in
  let foo_with_bar_baz = Baz.set foo_with_bar (Some "Test String") in

  (* Get extensions *)
  let open Ocaml_protoc_plugin.Result in
  Bar.get foo_with_bar >>= fun bar ->
  Baz.get foo_with_baz >>= fun baz ->
  assert (bar = Some 42);
  assert (baz = Some "Test String");
  Bar.get foo_with_bar_baz >>= fun bar' ->
  Baz.get foo_with_bar_baz >>= fun baz' ->
  assert (bar' = Some 42);
  assert (baz' = Some "Test String");
  return ()

Extensions are replaced by proto3 Any type, and use is discouraged.

Proto3 Any type

No special handling of any type is supported, as Ocaml does not allow for runtime types, so any type must be handled manually by serializing and deserializing the embedded message.

Imported protofiles

The generated code assumes that imported modules (generated from proto files) are available in the compilation scope. If the modules generated from imported protofiles resides in different a different scope (e.g. is compiled with wrapped true, they need to be made available by adding parameter open=<module name> to make the modules available for the compilation.

Google Well know types

Protobuf distributes a set of Well-Known types. ocaml-protoc-plugin installs compiled versions of these. These can be used by linking with the package ocaml-protoc-plugin.google_types, and adding option open=Google_types to the list of parameters

The distributed google types are compiled using default parameters, i.e. without any ppx annotations.

If you want to change this, or add type annotations, you can copy the dune from the distribution to your own project, and make alterations there. See the echo_deriving example on how to do this.

Example

test.proto

syntax = "proto3";
message Address {
  enum Planet {
    Earth = 0; Mars  = 1; Pluto = 2;
  }
  string street = 1;
  uint64 number = 2;
  Planet planet = 3;
}

message Person {
  uint64 id       = 1;
  string name     = 2;
  Address address = 3;
}

$ protoc --ocaml_out=. test.proto

Generates a file test.ml with the following signature:

module Address : sig
  module rec Planet : sig
    type t = Earth | Mars | Pluto
    val to_int: t -> int
    val from_int: int -> t Protobuf.Deserialize.result
  end
  val name': unit -> string
  type t = {
    street: string;
    number: int;
    planet: Planet.t;
  }
  val make: ?street:string ?number:int ?planet:Planet.t -> unit -> t
  val to_proto: t -> Protobuf.Writer.t
  val from_proto: Protobuf.Reader.t -> (t, Protobuf.Deserialize.error) result
end
module Person : sig
  val name': unit -> string
  type t = {
    id: int;
    name: string;
    address: Address.t option;
  }
  val make: ?id:int ?name:string ?planet:Address.t -> unit -> t
  val to_proto: t -> Protobuf.Writer.t
  val from_proto: Protobuf.Reader.t -> (t, Protobuf.Deserialize.error) result
end = struct

Note that if test.proto had a package declaration such as package testing, the modules Address and Person listed above would be defined as sub-modules of a top-level module Testing.

The function make allows the user to create message without specifying all (or any) fields. Using this function will allow users to add fields to message later without needing to modify any code, as new fields will be set to default values.

Protobuf.Reader and Protobuf.Writer are used then reading or writing protobuf binary format. Below is an example on how to decode a message and how to read a message.

let string_of_planet = function
  | Address.Earth -> "earth"
  | Mars -> "mars"
  | Pluto -> "pluto"
in

let read_person binary_message =
  let reader = Protobuf.Reader.create binary_message in
  match Person.from_proto reader in
  | Ok Person.{ id; name; address = Some Address { street; number; planet } } ->
    Printf.printf "P: %d %s - %s %s %d\n" id name (string_of_planet planet) street number
  | Ok Person.{ id; name; address = None } ->
    Printf.printf "P: %d %s - Address unknown\n" id name
  | Error _ -> failwith "Could not decode"

More examples can be found under examples